[llvm] [VPlan] Extract reverse operation for reverse accesses (PR #146525)

Mel Chen via llvm-commits llvm-commits at lists.llvm.org
Wed Aug 27 01:49:43 PDT 2025


https://github.com/Mel-Chen updated https://github.com/llvm/llvm-project/pull/146525

>From 7b22b76c860c46c1ba333bdcb12f34f7bd1e9ad7 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 1 Jul 2025 06:11:39 -0700
Subject: [PATCH 01/10] [VPlan] Extract reverse operation for reverse accesses

---
 .../Target/RISCV/RISCVTargetTransformInfo.cpp |  6 ++
 .../Transforms/Vectorize/LoopVectorize.cpp    |  4 +
 llvm/lib/Transforms/Vectorize/VPlan.h         |  2 +
 .../Transforms/Vectorize/VPlanAnalysis.cpp    |  1 +
 .../lib/Transforms/Vectorize/VPlanRecipes.cpp | 47 +++++-------
 .../Transforms/Vectorize/VPlanTransforms.cpp  | 53 +++++++++++++
 .../Transforms/Vectorize/VPlanTransforms.h    | 14 ++++
 .../AArch64/sve-vector-reverse-mask4.ll       |  2 +-
 .../AArch64/vector-reverse-mask4.ll           |  2 +-
 .../PowerPC/optimal-epilog-vectorization.ll   | 28 +++----
 .../RISCV/riscv-vector-reverse.ll             | 40 +++++-----
 .../RISCV/tail-folding-reverse-load-store.ll  | 10 +--
 .../RISCV/tail-folding-uniform-store.ll       | 36 ++++-----
 .../RISCV/vplan-riscv-vector-reverse.ll       |  8 +-
 .../LoopVectorize/X86/masked_load_store.ll    | 76 +++++++++----------
 .../interleave-with-i65-induction.ll          |  2 +-
 .../LoopVectorize/iv-select-cmp-decreasing.ll | 12 +--
 .../LoopVectorize/reverse_induction.ll        | 10 +--
 .../single-early-exit-interleave.ll           | 24 +++---
 .../vplan-sink-scalars-and-merge.ll           |  3 +-
 20 files changed, 226 insertions(+), 154 deletions(-)

diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index c707fb110b10c..7bd32ffe18b64 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -1523,6 +1523,12 @@ RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
                           cast<VectorType>(ICA.getArgTypes()[0]), {}, CostKind,
                           0, cast<VectorType>(ICA.getReturnType()));
   }
+  case Intrinsic::experimental_vp_reverse: {
+    return getShuffleCost(TTI::SK_Reverse,
+                          cast<VectorType>(ICA.getReturnType()),
+                          cast<VectorType>(ICA.getArgTypes()[0]), {}, CostKind,
+                          0, cast<VectorType>(ICA.getReturnType()));
+  }
   case Intrinsic::fptoui_sat:
   case Intrinsic::fptosi_sat: {
     InstructionCost Cost = 0;
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 6317bc3c20e25..8bc4a12e0299d 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -8747,6 +8747,10 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
   // bring the VPlan to its final state.
   // ---------------------------------------------------------------------------
 
+  // Adjust the result of reverse memory accesses.
+  VPlanTransforms::runPass(VPlanTransforms::adjustRecipesForReverseAccesses,
+                           *Plan);
+
   // Adjust the recipes for any inloop reductions.
   adjustRecipesForReductions(Plan, RecipeBuilder, Range.Start);
 
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 33bcb49b81740..e7bb052aa8d60 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1021,6 +1021,8 @@ class LLVM_ABI_FOR_TEST VPInstruction : public VPRecipeWithIRFlags,
     // It produces the lane index across all unrolled iterations. Unrolling will
     // add all copies of its original operand as additional operands.
     FirstActiveLane,
+    // Returns a reversed vector for the operand.
+    Reverse,
 
     // The opcodes below are used for VPInstructionWithType.
     //
diff --git a/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp b/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
index 747c6623aa22a..2680d01b77ed1 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
@@ -129,6 +129,7 @@ Type *VPTypeAnalysis::inferScalarTypeForRecipe(const VPInstruction *R) {
   case VPInstruction::Broadcast:
   case VPInstruction::PtrAdd:
   case VPInstruction::WidePtrAdd:
+  case VPInstruction::Reverse:
     // Return the type based on first operand.
     return inferScalarType(R->getOperand(0));
   case VPInstruction::BranchOnCond:
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 86834ab1240c1..2a669bbd32ac7 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -467,6 +467,7 @@ unsigned VPInstruction::getNumOperandsForOpcode(unsigned Opcode) {
   case VPInstruction::ExtractPenultimateElement:
   case VPInstruction::FirstActiveLane:
   case VPInstruction::Not:
+  case VPInstruction::Reverse:
     return 1;
   case Instruction::ICmp:
   case Instruction::FCmp:
@@ -942,6 +943,9 @@ Value *VPInstruction::generate(VPTransformState &State) {
   }
   case VPInstruction::ResumeForEpilogue:
     return State.get(getOperand(0), true);
+  case VPInstruction::Reverse: {
+    return Builder.CreateVectorReverse(State.get(getOperand(0)), "reverse");
+  }
   default:
     llvm_unreachable("Unsupported opcode for instruction");
   }
@@ -1104,6 +1108,13 @@ InstructionCost VPInstruction::computeCost(ElementCount VF,
                                   I32Ty, {Arg0Ty, I32Ty, I1Ty});
     return Ctx.TTI.getIntrinsicInstrCost(Attrs, Ctx.CostKind);
   }
+  case VPInstruction::Reverse: {
+    assert(VF.isVector() && "Reverse operation must be vector type");
+    Type *VectorTy = toVectorTy(Ctx.Types.inferScalarType(this), VF);
+    return Ctx.TTI.getShuffleCost(
+        TargetTransformInfo::SK_Reverse, cast<VectorType>(VectorTy),
+        cast<VectorType>(VectorTy), {}, Ctx.CostKind, 0);
+  }
   case VPInstruction::ExtractLastElement: {
     // Add on the cost of extracting the element.
     auto *VecTy = toVectorTy(Ctx.Types.inferScalarType(getOperand(0)), VF);
@@ -1208,6 +1219,7 @@ bool VPInstruction::opcodeMayReadOrWriteFromMemory() const {
   case VPInstruction::WidePtrAdd:
   case VPInstruction::StepVector:
   case VPInstruction::ReductionStartVector:
+  case VPInstruction::Reverse:
   case VPInstruction::VScale:
     return false;
   default:
@@ -1371,6 +1383,9 @@ void VPInstruction::print(raw_ostream &O, const Twine &Indent,
   case VPInstruction::ResumeForEpilogue:
     O << "resume-for-epilogue";
     break;
+  case VPInstruction::Reverse:
+    O << "reverse";
+    break;
   default:
     O << Instruction::getOpcodeName(getOpcode());
   }
@@ -3271,12 +3286,7 @@ InstructionCost VPWidenMemoryRecipe::computeCost(ElementCount VF,
     Cost += Ctx.TTI.getMemoryOpCost(Opcode, Ty, Alignment, AS, Ctx.CostKind,
                                     OpInfo, &Ingredient);
   }
-  if (!Reverse)
-    return Cost;
-
-  return Cost += Ctx.TTI.getShuffleCost(
-             TargetTransformInfo::SK_Reverse, cast<VectorType>(Ty),
-             cast<VectorType>(Ty), {}, Ctx.CostKind, 0);
+  return Cost;
 }
 
 void VPWidenLoadRecipe::execute(VPTransformState &State) {
@@ -3308,8 +3318,6 @@ void VPWidenLoadRecipe::execute(VPTransformState &State) {
     NewLI = Builder.CreateAlignedLoad(DataTy, Addr, Alignment, "wide.load");
   }
   applyMetadata(*cast<Instruction>(NewLI));
-  if (Reverse)
-    NewLI = Builder.CreateVectorReverse(NewLI, "reverse");
   State.set(this, NewLI);
 }
 
@@ -3365,8 +3373,6 @@ void VPWidenLoadEVLRecipe::execute(VPTransformState &State) {
       0, Attribute::getWithAlignment(NewLI->getContext(), Alignment));
   applyMetadata(*NewLI);
   Instruction *Res = NewLI;
-  if (isReverse())
-    Res = createReverseEVL(Builder, Res, EVL, "vp.reverse");
   State.set(this, Res);
 }
 
@@ -3385,12 +3391,8 @@ InstructionCost VPWidenLoadEVLRecipe::computeCost(ElementCount VF,
   unsigned AS = getLoadStoreAddressSpace(&Ingredient);
   InstructionCost Cost = Ctx.TTI.getMaskedMemoryOpCost(
       Instruction::Load, Ty, Alignment, AS, Ctx.CostKind);
-  if (!Reverse)
-    return Cost;
 
-  return Cost + Ctx.TTI.getShuffleCost(
-                    TargetTransformInfo::SK_Reverse, cast<VectorType>(Ty),
-                    cast<VectorType>(Ty), {}, Ctx.CostKind, 0);
+  return Cost;
 }
 
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
@@ -3420,13 +3422,6 @@ void VPWidenStoreRecipe::execute(VPTransformState &State) {
   }
 
   Value *StoredVal = State.get(StoredVPValue);
-  if (isReverse()) {
-    // If we store to reverse consecutive memory locations, then we need
-    // to reverse the order of elements in the stored value.
-    StoredVal = Builder.CreateVectorReverse(StoredVal, "reverse");
-    // We don't want to update the value in the map as it might be used in
-    // another expression. So don't call resetVectorValue(StoredVal).
-  }
   Value *Addr = State.get(getAddr(), /*IsScalar*/ !CreateScatter);
   Instruction *NewSI = nullptr;
   if (CreateScatter)
@@ -3456,8 +3451,6 @@ void VPWidenStoreEVLRecipe::execute(VPTransformState &State) {
   CallInst *NewSI = nullptr;
   Value *StoredVal = State.get(StoredValue);
   Value *EVL = State.get(getEVL(), VPLane(0));
-  if (isReverse())
-    StoredVal = createReverseEVL(Builder, StoredVal, EVL, "vp.reverse");
   Value *Mask = nullptr;
   if (VPValue *VPMask = getMask()) {
     Mask = State.get(VPMask);
@@ -3496,12 +3489,8 @@ InstructionCost VPWidenStoreEVLRecipe::computeCost(ElementCount VF,
   unsigned AS = getLoadStoreAddressSpace(&Ingredient);
   InstructionCost Cost = Ctx.TTI.getMaskedMemoryOpCost(
       Instruction::Store, Ty, Alignment, AS, Ctx.CostKind);
-  if (!Reverse)
-    return Cost;
 
-  return Cost + Ctx.TTI.getShuffleCost(
-                    TargetTransformInfo::SK_Reverse, cast<VectorType>(Ty),
-                    cast<VectorType>(Ty), {}, Ctx.CostKind, 0);
+  return Cost;
 }
 
 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index d32d2a9ad11f7..8c8b2c54769e3 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -2227,6 +2227,27 @@ static VPRecipeBase *optimizeMaskToEVL(VPValue *HeaderMask,
       .Default([&](VPRecipeBase *R) { return nullptr; });
 }
 
+static void convertToEVLReverse(VPlan &Plan, VPTypeAnalysis &TypeInfo,
+                                VPValue &AllOneMask, VPValue &EVL) {
+  for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(
+           vp_depth_first_shallow(Plan.getVectorLoopRegion()->getEntry()))) {
+    for (VPRecipeBase &R : make_early_inc_range(reverse(*VPBB))) {
+      auto *VPI = dyn_cast<VPInstruction>(&R);
+      if (!VPI || VPI->getOpcode() != VPInstruction::Reverse)
+        continue;
+
+      SmallVector<VPValue *> Ops(VPI->operands());
+      Ops.append({&AllOneMask, &EVL});
+      auto *NewReverse = new VPWidenIntrinsicRecipe(
+          Intrinsic::experimental_vp_reverse, Ops,
+          TypeInfo.inferScalarType(VPI), VPI->getDebugLoc());
+      NewReverse->insertBefore(VPI);
+      VPI->replaceAllUsesWith(NewReverse);
+      VPI->eraseFromParent();
+    }
+  }
+}
+
 /// Replace recipes with their EVL variants.
 static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
   VPTypeAnalysis TypeInfo(Plan);
@@ -2340,6 +2361,7 @@ static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
     }
     ToErase.push_back(CurRecipe);
   }
+  convertToEVLReverse(Plan, TypeInfo, *AllOneMask, EVL);
   // Remove dead EVL mask.
   if (EVLMask->getNumUsers() == 0)
     ToErase.push_back(EVLMask->getDefiningRecipe());
@@ -3798,3 +3820,34 @@ void VPlanTransforms::addBranchWeightToMiddleTerminator(
       MDB.createBranchWeights({1, VectorStep - 1}, /*IsExpected=*/false);
   MiddleTerm->addMetadata(LLVMContext::MD_prof, BranchWeights);
 }
+
+void VPlanTransforms::adjustRecipesForReverseAccesses(VPlan &Plan) {
+  if (Plan.hasScalarVFOnly())
+    return;
+
+  for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(
+           vp_depth_first_deep(Plan.getVectorLoopRegion()))) {
+    for (VPRecipeBase &R : *VPBB) {
+      auto *MemR = dyn_cast<VPWidenMemoryRecipe>(&R);
+      if (!MemR || !MemR->isReverse())
+        continue;
+
+      if (auto *L = dyn_cast<VPWidenLoadRecipe>(MemR)) {
+        auto *Reverse =
+            new VPInstruction(VPInstruction::Reverse, {L}, L->getDebugLoc());
+        Reverse->insertAfter(L);
+        L->replaceAllUsesWith(Reverse);
+        Reverse->setOperand(0, L);
+        continue;
+      }
+
+      if (auto *S = dyn_cast<VPWidenStoreRecipe>(MemR)) {
+        VPValue *StoredVal = S->getStoredValue();
+        auto *Reverse = new VPInstruction(VPInstruction::Reverse, {StoredVal},
+                                          S->getDebugLoc());
+        Reverse->insertBefore(S);
+        S->setOperand(1, Reverse);
+      }
+    }
+  }
+}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
index 9cf62a35ae36b..173aa6066ab83 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
@@ -347,6 +347,20 @@ struct VPlanTransforms {
   static void
   addBranchWeightToMiddleTerminator(VPlan &Plan, ElementCount VF,
                                     std::optional<unsigned> VScaleForTuning);
+
+  /// Add reverse recipes for reverse memory accesses.
+  /// For reverse loads, transform
+  ///   WIDEN ir<%L> = load vp<%addr>
+  /// into
+  ///   WIDEN ir<%L> = load vp<%addr>
+  ///   EMIT   vp<%RevL> = reverse ir<%L>
+  ///
+  /// For reverse stores, transform
+  ///   WIDEN store vp<%addr>, ir<%SVal>
+  /// into
+  ///   EMIT   vp<%RevS> = reverse ir<%SVal>
+  ///   WIDEN  store vp<%addr>, vp<%RevS>
+  static void adjustRecipesForReverseAccesses(VPlan &Plan);
 };
 
 } // namespace llvm
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-vector-reverse-mask4.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-vector-reverse-mask4.ll
index 9485d827ced40..c838c63545341 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-vector-reverse-mask4.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-vector-reverse-mask4.ll
@@ -22,8 +22,8 @@ define void @vector_reverse_mask_nxv4i1(ptr %a, ptr %cond, i64 %N) #0 {
 ; CHECK: %[[WIDEMSKLOAD:.*]] = call <vscale x 4 x double> @llvm.masked.load.nxv4f64.p0(ptr %{{.*}}, i32 8, <vscale x 4 x i1> %[[REVERSE6]], <vscale x 4 x double> poison)
 ; CHECK: %[[REVERSE7:.*]] = call <vscale x 4 x double> @llvm.vector.reverse.nxv4f64(<vscale x 4 x double> %[[WIDEMSKLOAD]])
 ; CHECK: %[[FADD:.*]] = fadd <vscale x 4 x double> %[[REVERSE7]]
-; CHECK: %[[REVERSE9:.*]] = call <vscale x 4 x i1> @llvm.vector.reverse.nxv4i1(<vscale x 4 x i1> %{{.*}})
 ; CHECK: %[[REVERSE8:.*]] = call <vscale x 4 x double> @llvm.vector.reverse.nxv4f64(<vscale x 4 x double> %[[FADD]])
+; CHECK: %[[REVERSE9:.*]] = call <vscale x 4 x i1> @llvm.vector.reverse.nxv4i1(<vscale x 4 x i1> %{{.*}})
 ; CHECK: call void @llvm.masked.store.nxv4f64.p0(<vscale x 4 x double> %[[REVERSE8]], ptr %{{.*}}, i32 8, <vscale x 4 x i1> %[[REVERSE9]]
 
 entry:
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/vector-reverse-mask4.ll b/llvm/test/Transforms/LoopVectorize/AArch64/vector-reverse-mask4.ll
index 1dd49ecf85b81..d6f619cce54a0 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/vector-reverse-mask4.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/vector-reverse-mask4.ll
@@ -37,8 +37,8 @@ define void @vector_reverse_mask_v4i1(ptr noalias %a, ptr noalias %cond, i64 %N)
 ; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[TMP2]], i64 -24
 ; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[TMP2]], i64 -56
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x double>, ptr [[TMP3]], align 8
-; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x double> [[WIDE_LOAD]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[WIDE_LOAD1:%.*]] = load <4 x double>, ptr [[TMP4]], align 8
+; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x double> [[WIDE_LOAD]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[REVERSE2:%.*]] = shufflevector <4 x double> [[WIDE_LOAD1]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[TMP5:%.*]] = fcmp une <4 x double> [[REVERSE]], zeroinitializer
 ; CHECK-NEXT:    [[TMP6:%.*]] = fcmp une <4 x double> [[REVERSE2]], zeroinitializer
diff --git a/llvm/test/Transforms/LoopVectorize/PowerPC/optimal-epilog-vectorization.ll b/llvm/test/Transforms/LoopVectorize/PowerPC/optimal-epilog-vectorization.ll
index 4d146244c9ab1..4957699ca83d5 100644
--- a/llvm/test/Transforms/LoopVectorize/PowerPC/optimal-epilog-vectorization.ll
+++ b/llvm/test/Transforms/LoopVectorize/PowerPC/optimal-epilog-vectorization.ll
@@ -339,20 +339,20 @@ define dso_local signext i32 @f2(ptr noalias %A, ptr noalias %B, i32 signext %n)
 ; VF-TWO-CHECK-NEXT:    [[TMP70:%.*]] = getelementptr inbounds float, ptr [[TMP48]], i32 -28
 ; VF-TWO-CHECK-NEXT:    [[TMP71:%.*]] = getelementptr inbounds float, ptr [[TMP70]], i32 -3
 ; VF-TWO-CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP57]], align 4
-; VF-TWO-CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x float> [[WIDE_LOAD]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[WIDE_LOAD2:%.*]] = load <4 x float>, ptr [[TMP59]], align 4
-; VF-TWO-CHECK-NEXT:    [[REVERSE3:%.*]] = shufflevector <4 x float> [[WIDE_LOAD2]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x float>, ptr [[TMP61]], align 4
-; VF-TWO-CHECK-NEXT:    [[REVERSE5:%.*]] = shufflevector <4 x float> [[WIDE_LOAD4]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x float>, ptr [[TMP63]], align 4
-; VF-TWO-CHECK-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x float> [[WIDE_LOAD6]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x float>, ptr [[TMP65]], align 4
-; VF-TWO-CHECK-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x float> [[WIDE_LOAD8]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[WIDE_LOAD10:%.*]] = load <4 x float>, ptr [[TMP67]], align 4
-; VF-TWO-CHECK-NEXT:    [[REVERSE11:%.*]] = shufflevector <4 x float> [[WIDE_LOAD10]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[WIDE_LOAD12:%.*]] = load <4 x float>, ptr [[TMP69]], align 4
-; VF-TWO-CHECK-NEXT:    [[REVERSE13:%.*]] = shufflevector <4 x float> [[WIDE_LOAD12]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[WIDE_LOAD14:%.*]] = load <4 x float>, ptr [[TMP71]], align 4
+; VF-TWO-CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x float> [[WIDE_LOAD]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-TWO-CHECK-NEXT:    [[REVERSE3:%.*]] = shufflevector <4 x float> [[WIDE_LOAD2]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-TWO-CHECK-NEXT:    [[REVERSE5:%.*]] = shufflevector <4 x float> [[WIDE_LOAD4]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-TWO-CHECK-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x float> [[WIDE_LOAD6]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-TWO-CHECK-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x float> [[WIDE_LOAD8]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-TWO-CHECK-NEXT:    [[REVERSE11:%.*]] = shufflevector <4 x float> [[WIDE_LOAD10]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-TWO-CHECK-NEXT:    [[REVERSE13:%.*]] = shufflevector <4 x float> [[WIDE_LOAD12]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[REVERSE15:%.*]] = shufflevector <4 x float> [[WIDE_LOAD14]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-TWO-CHECK-NEXT:    [[TMP72:%.*]] = fadd fast <4 x float> [[REVERSE]], splat (float 1.000000e+00)
 ; VF-TWO-CHECK-NEXT:    [[TMP73:%.*]] = fadd fast <4 x float> [[REVERSE3]], splat (float 1.000000e+00)
@@ -492,20 +492,20 @@ define dso_local signext i32 @f2(ptr noalias %A, ptr noalias %B, i32 signext %n)
 ; VF-FOUR-CHECK-NEXT:    [[TMP70:%.*]] = getelementptr inbounds float, ptr [[TMP48]], i32 -28
 ; VF-FOUR-CHECK-NEXT:    [[TMP71:%.*]] = getelementptr inbounds float, ptr [[TMP70]], i32 -3
 ; VF-FOUR-CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP57]], align 4
-; VF-FOUR-CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x float> [[WIDE_LOAD]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[WIDE_LOAD2:%.*]] = load <4 x float>, ptr [[TMP59]], align 4
-; VF-FOUR-CHECK-NEXT:    [[REVERSE3:%.*]] = shufflevector <4 x float> [[WIDE_LOAD2]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x float>, ptr [[TMP61]], align 4
-; VF-FOUR-CHECK-NEXT:    [[REVERSE5:%.*]] = shufflevector <4 x float> [[WIDE_LOAD4]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x float>, ptr [[TMP63]], align 4
-; VF-FOUR-CHECK-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x float> [[WIDE_LOAD6]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x float>, ptr [[TMP65]], align 4
-; VF-FOUR-CHECK-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x float> [[WIDE_LOAD8]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[WIDE_LOAD10:%.*]] = load <4 x float>, ptr [[TMP67]], align 4
-; VF-FOUR-CHECK-NEXT:    [[REVERSE11:%.*]] = shufflevector <4 x float> [[WIDE_LOAD10]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[WIDE_LOAD12:%.*]] = load <4 x float>, ptr [[TMP69]], align 4
-; VF-FOUR-CHECK-NEXT:    [[REVERSE13:%.*]] = shufflevector <4 x float> [[WIDE_LOAD12]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[WIDE_LOAD14:%.*]] = load <4 x float>, ptr [[TMP71]], align 4
+; VF-FOUR-CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x float> [[WIDE_LOAD]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-FOUR-CHECK-NEXT:    [[REVERSE3:%.*]] = shufflevector <4 x float> [[WIDE_LOAD2]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-FOUR-CHECK-NEXT:    [[REVERSE5:%.*]] = shufflevector <4 x float> [[WIDE_LOAD4]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-FOUR-CHECK-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x float> [[WIDE_LOAD6]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-FOUR-CHECK-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x float> [[WIDE_LOAD8]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-FOUR-CHECK-NEXT:    [[REVERSE11:%.*]] = shufflevector <4 x float> [[WIDE_LOAD10]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF-FOUR-CHECK-NEXT:    [[REVERSE13:%.*]] = shufflevector <4 x float> [[WIDE_LOAD12]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[REVERSE15:%.*]] = shufflevector <4 x float> [[WIDE_LOAD14]], <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF-FOUR-CHECK-NEXT:    [[TMP72:%.*]] = fadd fast <4 x float> [[REVERSE]], splat (float 1.000000e+00)
 ; VF-FOUR-CHECK-NEXT:    [[TMP73:%.*]] = fadd fast <4 x float> [[REVERSE3]], splat (float 1.000000e+00)
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/riscv-vector-reverse.ll b/llvm/test/Transforms/LoopVectorize/RISCV/riscv-vector-reverse.ll
index 8272e6d82295c..bb5c4c127c0d6 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/riscv-vector-reverse.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/riscv-vector-reverse.ll
@@ -39,13 +39,13 @@ define void @vector_reverse_i32(ptr noalias %A, ptr noalias %B) {
 ; RV64-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP19]])
 ; RV64-NEXT:    [[TMP14:%.*]] = add <vscale x 4 x i32> [[REVERSE]], splat (i32 1)
 ; RV64-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP7]]
+; RV64-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[TMP14]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP19]])
 ; RV64-NEXT:    [[TMP16:%.*]] = zext i32 [[TMP19]] to i64
 ; RV64-NEXT:    [[TMP25:%.*]] = mul i64 0, [[TMP16]]
 ; RV64-NEXT:    [[TMP17:%.*]] = sub i64 [[TMP16]], 1
 ; RV64-NEXT:    [[TMP18:%.*]] = mul i64 -1, [[TMP17]]
 ; RV64-NEXT:    [[TMP20:%.*]] = getelementptr i32, ptr [[TMP15]], i64 [[TMP25]]
 ; RV64-NEXT:    [[TMP21:%.*]] = getelementptr i32, ptr [[TMP20]], i64 [[TMP18]]
-; RV64-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[TMP14]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP19]])
 ; RV64-NEXT:    call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_REVERSE1]], ptr align 4 [[TMP21]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP19]])
 ; RV64-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP19]] to i64
 ; RV64-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP22]], [[INDEX]]
@@ -80,12 +80,12 @@ define void @vector_reverse_i32(ptr noalias %A, ptr noalias %B) {
 ; RV32-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP9]])
 ; RV32-NEXT:    [[TMP15:%.*]] = add <vscale x 4 x i32> [[REVERSE]], splat (i32 1)
 ; RV32-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP7]]
+; RV32-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[TMP15]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP9]])
 ; RV32-NEXT:    [[TMP17:%.*]] = mul i32 0, [[TMP9]]
 ; RV32-NEXT:    [[TMP19:%.*]] = sub i32 [[TMP9]], 1
 ; RV32-NEXT:    [[TMP20:%.*]] = mul i32 -1, [[TMP19]]
 ; RV32-NEXT:    [[TMP18:%.*]] = getelementptr i32, ptr [[TMP16]], i32 [[TMP17]]
 ; RV32-NEXT:    [[TMP22:%.*]] = getelementptr i32, ptr [[TMP18]], i32 [[TMP20]]
-; RV32-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[TMP15]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP9]])
 ; RV32-NEXT:    call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_REVERSE1]], ptr align 4 [[TMP22]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP9]])
 ; RV32-NEXT:    [[TMP23:%.*]] = zext i32 [[TMP9]] to i64
 ; RV32-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP23]], [[INDEX]]
@@ -129,8 +129,8 @@ define void @vector_reverse_i32(ptr noalias %A, ptr noalias %B) {
 ; RV64-UF2-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i32, ptr [[TMP9]], i64 [[TMP15]]
 ; RV64-UF2-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[TMP18]], i64 [[TMP17]]
 ; RV64-UF2-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP14]], align 4
-; RV64-UF2-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[WIDE_LOAD]])
 ; RV64-UF2-NEXT:    [[WIDE_LOAD1:%.*]] = load <vscale x 4 x i32>, ptr [[TMP19]], align 4
+; RV64-UF2-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[WIDE_LOAD]])
 ; RV64-UF2-NEXT:    [[REVERSE2:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[WIDE_LOAD1]])
 ; RV64-UF2-NEXT:    [[TMP20:%.*]] = add <vscale x 4 x i32> [[REVERSE]], splat (i32 1)
 ; RV64-UF2-NEXT:    [[TMP21:%.*]] = add <vscale x 4 x i32> [[REVERSE2]], splat (i32 1)
@@ -146,8 +146,8 @@ define void @vector_reverse_i32(ptr noalias %A, ptr noalias %B) {
 ; RV64-UF2-NEXT:    [[TMP31:%.*]] = getelementptr inbounds i32, ptr [[TMP22]], i64 [[TMP28]]
 ; RV64-UF2-NEXT:    [[TMP32:%.*]] = getelementptr inbounds i32, ptr [[TMP31]], i64 [[TMP30]]
 ; RV64-UF2-NEXT:    [[REVERSE3:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[TMP20]])
-; RV64-UF2-NEXT:    store <vscale x 4 x i32> [[REVERSE3]], ptr [[TMP27]], align 4
 ; RV64-UF2-NEXT:    [[REVERSE4:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[TMP21]])
+; RV64-UF2-NEXT:    store <vscale x 4 x i32> [[REVERSE3]], ptr [[TMP27]], align 4
 ; RV64-UF2-NEXT:    store <vscale x 4 x i32> [[REVERSE4]], ptr [[TMP32]], align 4
 ; RV64-UF2-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]
 ; RV64-UF2-NEXT:    [[TMP34:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[TMP7]]
@@ -230,13 +230,13 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV64-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP20]])
 ; RV64-NEXT:    [[TMP29:%.*]] = add <vscale x 4 x i32> [[REVERSE]], splat (i32 1)
 ; RV64-NEXT:    [[TMP30:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP22]]
+; RV64-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[TMP29]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP20]])
 ; RV64-NEXT:    [[TMP39:%.*]] = zext i32 [[TMP20]] to i64
 ; RV64-NEXT:    [[TMP31:%.*]] = mul i64 0, [[TMP39]]
 ; RV64-NEXT:    [[TMP32:%.*]] = sub i64 [[TMP39]], 1
 ; RV64-NEXT:    [[TMP33:%.*]] = mul i64 -1, [[TMP32]]
 ; RV64-NEXT:    [[TMP34:%.*]] = getelementptr i32, ptr [[TMP30]], i64 [[TMP31]]
 ; RV64-NEXT:    [[TMP35:%.*]] = getelementptr i32, ptr [[TMP34]], i64 [[TMP33]]
-; RV64-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[TMP29]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP20]])
 ; RV64-NEXT:    call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_REVERSE3]], ptr align 4 [[TMP35]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP20]])
 ; RV64-NEXT:    [[TMP36:%.*]] = zext i32 [[TMP20]] to i64
 ; RV64-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP36]], [[INDEX]]
@@ -247,7 +247,7 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV64-NEXT:    br label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]]
 ; RV64:       [[SCALAR_PH]]:
 ; RV64-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP0]], %[[FOR_BODY_PREHEADER]] ], [ [[TMP0]], %[[VECTOR_SCEVCHECK]] ], [ [[TMP0]], %[[VECTOR_MEMCHECK]] ]
-; RV64-NEXT:    [[BC_RESUME_VAL4:%.*]] = phi i32 [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_SCEVCHECK]] ], [ [[N]], %[[VECTOR_MEMCHECK]] ]
+; RV64-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_SCEVCHECK]] ], [ [[N]], %[[VECTOR_MEMCHECK]] ]
 ; RV64-NEXT:    br label %[[FOR_BODY:.*]]
 ; RV64:       [[FOR_COND_CLEANUP_LOOPEXIT]]:
 ; RV64-NEXT:    br label %[[FOR_COND_CLEANUP]]
@@ -292,12 +292,12 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV32-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP16]])
 ; RV32-NEXT:    [[TMP22:%.*]] = add <vscale x 4 x i32> [[REVERSE]], splat (i32 1)
 ; RV32-NEXT:    [[TMP23:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP14]]
+; RV32-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[TMP22]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP16]])
 ; RV32-NEXT:    [[TMP21:%.*]] = mul i32 0, [[TMP16]]
 ; RV32-NEXT:    [[TMP26:%.*]] = sub i32 [[TMP16]], 1
 ; RV32-NEXT:    [[TMP27:%.*]] = mul i32 -1, [[TMP26]]
 ; RV32-NEXT:    [[TMP24:%.*]] = getelementptr i32, ptr [[TMP23]], i32 [[TMP21]]
 ; RV32-NEXT:    [[TMP25:%.*]] = getelementptr i32, ptr [[TMP24]], i32 [[TMP27]]
-; RV32-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[TMP22]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP16]])
 ; RV32-NEXT:    call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_REVERSE3]], ptr align 4 [[TMP25]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP16]])
 ; RV32-NEXT:    [[TMP29:%.*]] = zext i32 [[TMP16]] to i64
 ; RV32-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP29]], [[INDEX]]
@@ -308,7 +308,7 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV32-NEXT:    br label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]]
 ; RV32:       [[SCALAR_PH]]:
 ; RV32-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP0]], %[[FOR_BODY_PREHEADER]] ], [ [[TMP0]], %[[VECTOR_MEMCHECK]] ]
-; RV32-NEXT:    [[BC_RESUME_VAL4:%.*]] = phi i32 [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_MEMCHECK]] ]
+; RV32-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_MEMCHECK]] ]
 ; RV32-NEXT:    br label %[[FOR_BODY:.*]]
 ; RV32:       [[FOR_COND_CLEANUP_LOOPEXIT]]:
 ; RV32-NEXT:    br label %[[FOR_COND_CLEANUP]]
@@ -377,8 +377,8 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV64-UF2-NEXT:    [[TMP33:%.*]] = getelementptr inbounds i32, ptr [[TMP24]], i64 [[TMP30]]
 ; RV64-UF2-NEXT:    [[TMP34:%.*]] = getelementptr inbounds i32, ptr [[TMP33]], i64 [[TMP32]]
 ; RV64-UF2-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP29]], align 4
-; RV64-UF2-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[WIDE_LOAD]])
 ; RV64-UF2-NEXT:    [[WIDE_LOAD4:%.*]] = load <vscale x 4 x i32>, ptr [[TMP34]], align 4
+; RV64-UF2-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[WIDE_LOAD]])
 ; RV64-UF2-NEXT:    [[REVERSE5:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[WIDE_LOAD4]])
 ; RV64-UF2-NEXT:    [[TMP35:%.*]] = add <vscale x 4 x i32> [[REVERSE]], splat (i32 1)
 ; RV64-UF2-NEXT:    [[TMP36:%.*]] = add <vscale x 4 x i32> [[REVERSE5]], splat (i32 1)
@@ -394,8 +394,8 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV64-UF2-NEXT:    [[TMP46:%.*]] = getelementptr inbounds i32, ptr [[TMP37]], i64 [[TMP43]]
 ; RV64-UF2-NEXT:    [[TMP47:%.*]] = getelementptr inbounds i32, ptr [[TMP46]], i64 [[TMP45]]
 ; RV64-UF2-NEXT:    [[REVERSE6:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[TMP35]])
-; RV64-UF2-NEXT:    store <vscale x 4 x i32> [[REVERSE6]], ptr [[TMP42]], align 4
 ; RV64-UF2-NEXT:    [[REVERSE7:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[TMP36]])
+; RV64-UF2-NEXT:    store <vscale x 4 x i32> [[REVERSE6]], ptr [[TMP42]], align 4
 ; RV64-UF2-NEXT:    store <vscale x 4 x i32> [[REVERSE7]], ptr [[TMP47]], align 4
 ; RV64-UF2-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP19]]
 ; RV64-UF2-NEXT:    [[TMP49:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[TMP20]]
@@ -491,13 +491,13 @@ define void @vector_reverse_f32(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV64-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP20]])
 ; RV64-NEXT:    [[TMP29:%.*]] = fadd <vscale x 4 x float> [[REVERSE]], splat (float 1.000000e+00)
 ; RV64-NEXT:    [[TMP30:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP22]]
+; RV64-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[TMP29]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP20]])
 ; RV64-NEXT:    [[TMP39:%.*]] = zext i32 [[TMP20]] to i64
 ; RV64-NEXT:    [[TMP31:%.*]] = mul i64 0, [[TMP39]]
 ; RV64-NEXT:    [[TMP32:%.*]] = sub i64 [[TMP39]], 1
 ; RV64-NEXT:    [[TMP33:%.*]] = mul i64 -1, [[TMP32]]
 ; RV64-NEXT:    [[TMP34:%.*]] = getelementptr float, ptr [[TMP30]], i64 [[TMP31]]
 ; RV64-NEXT:    [[TMP35:%.*]] = getelementptr float, ptr [[TMP34]], i64 [[TMP33]]
-; RV64-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[TMP29]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP20]])
 ; RV64-NEXT:    call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[VP_REVERSE3]], ptr align 4 [[TMP35]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP20]])
 ; RV64-NEXT:    [[TMP36:%.*]] = zext i32 [[TMP20]] to i64
 ; RV64-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP36]], [[INDEX]]
@@ -508,7 +508,7 @@ define void @vector_reverse_f32(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV64-NEXT:    br label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]]
 ; RV64:       [[SCALAR_PH]]:
 ; RV64-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP0]], %[[FOR_BODY_PREHEADER]] ], [ [[TMP0]], %[[VECTOR_SCEVCHECK]] ], [ [[TMP0]], %[[VECTOR_MEMCHECK]] ]
-; RV64-NEXT:    [[BC_RESUME_VAL4:%.*]] = phi i32 [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_SCEVCHECK]] ], [ [[N]], %[[VECTOR_MEMCHECK]] ]
+; RV64-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_SCEVCHECK]] ], [ [[N]], %[[VECTOR_MEMCHECK]] ]
 ; RV64-NEXT:    br label %[[FOR_BODY:.*]]
 ; RV64:       [[FOR_COND_CLEANUP_LOOPEXIT]]:
 ; RV64-NEXT:    br label %[[FOR_COND_CLEANUP]]
@@ -553,12 +553,12 @@ define void @vector_reverse_f32(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV32-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP16]])
 ; RV32-NEXT:    [[TMP22:%.*]] = fadd <vscale x 4 x float> [[REVERSE]], splat (float 1.000000e+00)
 ; RV32-NEXT:    [[TMP23:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP14]]
+; RV32-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[TMP22]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP16]])
 ; RV32-NEXT:    [[TMP21:%.*]] = mul i32 0, [[TMP16]]
 ; RV32-NEXT:    [[TMP26:%.*]] = sub i32 [[TMP16]], 1
 ; RV32-NEXT:    [[TMP27:%.*]] = mul i32 -1, [[TMP26]]
 ; RV32-NEXT:    [[TMP24:%.*]] = getelementptr float, ptr [[TMP23]], i32 [[TMP21]]
 ; RV32-NEXT:    [[TMP25:%.*]] = getelementptr float, ptr [[TMP24]], i32 [[TMP27]]
-; RV32-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[TMP22]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP16]])
 ; RV32-NEXT:    call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[VP_REVERSE3]], ptr align 4 [[TMP25]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP16]])
 ; RV32-NEXT:    [[TMP29:%.*]] = zext i32 [[TMP16]] to i64
 ; RV32-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP29]], [[INDEX]]
@@ -569,7 +569,7 @@ define void @vector_reverse_f32(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV32-NEXT:    br label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]]
 ; RV32:       [[SCALAR_PH]]:
 ; RV32-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP0]], %[[FOR_BODY_PREHEADER]] ], [ [[TMP0]], %[[VECTOR_MEMCHECK]] ]
-; RV32-NEXT:    [[BC_RESUME_VAL4:%.*]] = phi i32 [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_MEMCHECK]] ]
+; RV32-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[N]], %[[FOR_BODY_PREHEADER]] ], [ [[N]], %[[VECTOR_MEMCHECK]] ]
 ; RV32-NEXT:    br label %[[FOR_BODY:.*]]
 ; RV32:       [[FOR_COND_CLEANUP_LOOPEXIT]]:
 ; RV32-NEXT:    br label %[[FOR_COND_CLEANUP]]
@@ -638,8 +638,8 @@ define void @vector_reverse_f32(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV64-UF2-NEXT:    [[TMP33:%.*]] = getelementptr inbounds float, ptr [[TMP24]], i64 [[TMP30]]
 ; RV64-UF2-NEXT:    [[TMP34:%.*]] = getelementptr inbounds float, ptr [[TMP33]], i64 [[TMP32]]
 ; RV64-UF2-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x float>, ptr [[TMP29]], align 4
-; RV64-UF2-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD]])
 ; RV64-UF2-NEXT:    [[WIDE_LOAD4:%.*]] = load <vscale x 4 x float>, ptr [[TMP34]], align 4
+; RV64-UF2-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD]])
 ; RV64-UF2-NEXT:    [[REVERSE5:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD4]])
 ; RV64-UF2-NEXT:    [[TMP35:%.*]] = fadd <vscale x 4 x float> [[REVERSE]], splat (float 1.000000e+00)
 ; RV64-UF2-NEXT:    [[TMP36:%.*]] = fadd <vscale x 4 x float> [[REVERSE5]], splat (float 1.000000e+00)
@@ -655,8 +655,8 @@ define void @vector_reverse_f32(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; RV64-UF2-NEXT:    [[TMP46:%.*]] = getelementptr inbounds float, ptr [[TMP37]], i64 [[TMP43]]
 ; RV64-UF2-NEXT:    [[TMP47:%.*]] = getelementptr inbounds float, ptr [[TMP46]], i64 [[TMP45]]
 ; RV64-UF2-NEXT:    [[REVERSE6:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[TMP35]])
-; RV64-UF2-NEXT:    store <vscale x 4 x float> [[REVERSE6]], ptr [[TMP42]], align 4
 ; RV64-UF2-NEXT:    [[REVERSE7:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[TMP36]])
+; RV64-UF2-NEXT:    store <vscale x 4 x float> [[REVERSE6]], ptr [[TMP42]], align 4
 ; RV64-UF2-NEXT:    store <vscale x 4 x float> [[REVERSE7]], ptr [[TMP47]], align 4
 ; RV64-UF2-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP19]]
 ; RV64-UF2-NEXT:    [[TMP49:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[TMP20]]
@@ -724,13 +724,13 @@ define void @vector_reverse_f32_simplify(ptr noalias %A, ptr noalias %B) {
 ; RV64-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP19]])
 ; RV64-NEXT:    [[TMP14:%.*]] = fadd <vscale x 4 x float> [[REVERSE]], splat (float 1.000000e+00)
 ; RV64-NEXT:    [[TMP15:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP7]]
+; RV64-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[TMP14]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP19]])
 ; RV64-NEXT:    [[TMP16:%.*]] = zext i32 [[TMP19]] to i64
 ; RV64-NEXT:    [[TMP25:%.*]] = mul i64 0, [[TMP16]]
 ; RV64-NEXT:    [[TMP17:%.*]] = sub i64 [[TMP16]], 1
 ; RV64-NEXT:    [[TMP18:%.*]] = mul i64 -1, [[TMP17]]
 ; RV64-NEXT:    [[TMP20:%.*]] = getelementptr float, ptr [[TMP15]], i64 [[TMP25]]
 ; RV64-NEXT:    [[TMP21:%.*]] = getelementptr float, ptr [[TMP20]], i64 [[TMP18]]
-; RV64-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[TMP14]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP19]])
 ; RV64-NEXT:    call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[VP_REVERSE1]], ptr align 4 [[TMP21]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP19]])
 ; RV64-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP19]] to i64
 ; RV64-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP22]], [[INDEX]]
@@ -765,12 +765,12 @@ define void @vector_reverse_f32_simplify(ptr noalias %A, ptr noalias %B) {
 ; RV32-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP9]])
 ; RV32-NEXT:    [[TMP15:%.*]] = fadd <vscale x 4 x float> [[REVERSE]], splat (float 1.000000e+00)
 ; RV32-NEXT:    [[TMP16:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP7]]
+; RV32-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[TMP15]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP9]])
 ; RV32-NEXT:    [[TMP17:%.*]] = mul i32 0, [[TMP9]]
 ; RV32-NEXT:    [[TMP19:%.*]] = sub i32 [[TMP9]], 1
 ; RV32-NEXT:    [[TMP20:%.*]] = mul i32 -1, [[TMP19]]
 ; RV32-NEXT:    [[TMP18:%.*]] = getelementptr float, ptr [[TMP16]], i32 [[TMP17]]
 ; RV32-NEXT:    [[TMP22:%.*]] = getelementptr float, ptr [[TMP18]], i32 [[TMP20]]
-; RV32-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 4 x float> @llvm.experimental.vp.reverse.nxv4f32(<vscale x 4 x float> [[TMP15]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP9]])
 ; RV32-NEXT:    call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[VP_REVERSE1]], ptr align 4 [[TMP22]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP9]])
 ; RV32-NEXT:    [[TMP23:%.*]] = zext i32 [[TMP9]] to i64
 ; RV32-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP23]], [[INDEX]]
@@ -814,8 +814,8 @@ define void @vector_reverse_f32_simplify(ptr noalias %A, ptr noalias %B) {
 ; RV64-UF2-NEXT:    [[TMP18:%.*]] = getelementptr inbounds float, ptr [[TMP9]], i64 [[TMP15]]
 ; RV64-UF2-NEXT:    [[TMP19:%.*]] = getelementptr inbounds float, ptr [[TMP18]], i64 [[TMP17]]
 ; RV64-UF2-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 4 x float>, ptr [[TMP14]], align 4
-; RV64-UF2-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD]])
 ; RV64-UF2-NEXT:    [[WIDE_LOAD1:%.*]] = load <vscale x 4 x float>, ptr [[TMP19]], align 4
+; RV64-UF2-NEXT:    [[REVERSE:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD]])
 ; RV64-UF2-NEXT:    [[REVERSE2:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD1]])
 ; RV64-UF2-NEXT:    [[TMP20:%.*]] = fadd <vscale x 4 x float> [[REVERSE]], splat (float 1.000000e+00)
 ; RV64-UF2-NEXT:    [[TMP21:%.*]] = fadd <vscale x 4 x float> [[REVERSE2]], splat (float 1.000000e+00)
@@ -831,8 +831,8 @@ define void @vector_reverse_f32_simplify(ptr noalias %A, ptr noalias %B) {
 ; RV64-UF2-NEXT:    [[TMP31:%.*]] = getelementptr inbounds float, ptr [[TMP22]], i64 [[TMP28]]
 ; RV64-UF2-NEXT:    [[TMP32:%.*]] = getelementptr inbounds float, ptr [[TMP31]], i64 [[TMP30]]
 ; RV64-UF2-NEXT:    [[REVERSE3:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[TMP20]])
-; RV64-UF2-NEXT:    store <vscale x 4 x float> [[REVERSE3]], ptr [[TMP27]], align 4
 ; RV64-UF2-NEXT:    [[REVERSE4:%.*]] = call <vscale x 4 x float> @llvm.vector.reverse.nxv4f32(<vscale x 4 x float> [[TMP21]])
+; RV64-UF2-NEXT:    store <vscale x 4 x float> [[REVERSE3]], ptr [[TMP27]], align 4
 ; RV64-UF2-NEXT:    store <vscale x 4 x float> [[REVERSE4]], ptr [[TMP32]], align 4
 ; RV64-UF2-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]
 ; RV64-UF2-NEXT:    [[TMP34:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[TMP7]]
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-reverse-load-store.ll b/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-reverse-load-store.ll
index 908c06230b092..36dfd00e4c8aa 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-reverse-load-store.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-reverse-load-store.ll
@@ -29,13 +29,13 @@ define void @reverse_load_store(i64 %startval, ptr noalias %ptr, ptr noalias %pt
 ; IF-EVL-NEXT:    [[VP_OP_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP12]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[VP_REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[PTR2:%.*]], i64 [[TMP7]]
+; IF-EVL-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_REVERSE]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[TMP19:%.*]] = zext i32 [[TMP5]] to i64
 ; IF-EVL-NEXT:    [[TMP14:%.*]] = mul i64 0, [[TMP19]]
 ; IF-EVL-NEXT:    [[TMP23:%.*]] = sub i64 [[TMP19]], 1
 ; IF-EVL-NEXT:    [[TMP15:%.*]] = mul i64 -1, [[TMP23]]
 ; IF-EVL-NEXT:    [[TMP22:%.*]] = getelementptr i32, ptr [[TMP13]], i64 [[TMP14]]
 ; IF-EVL-NEXT:    [[TMP17:%.*]] = getelementptr i32, ptr [[TMP22]], i64 [[TMP15]]
-; IF-EVL-NEXT:    [[VP_REVERSE3:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_REVERSE]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_REVERSE3]], ptr align 4 [[TMP17]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP5]] to i64
 ; IF-EVL-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP20]], [[EVL_BASED_IV]]
@@ -164,13 +164,13 @@ define void @reverse_load_store_masked(i64 %startval, ptr noalias %ptr, ptr noal
 ; IF-EVL-NEXT:    [[VP_OP_LOAD4:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP20]], <vscale x 4 x i1> [[VP_REVERSE_MASK]], i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[VP_REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD4]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[TMP21:%.*]] = getelementptr i32, ptr [[PTR2:%.*]], i64 [[TMP11]]
+; IF-EVL-NEXT:    [[VP_REVERSE5:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_REVERSE]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[TMP27:%.*]] = zext i32 [[TMP5]] to i64
 ; IF-EVL-NEXT:    [[TMP22:%.*]] = mul i64 0, [[TMP27]]
 ; IF-EVL-NEXT:    [[TMP30:%.*]] = sub i64 [[TMP27]], 1
 ; IF-EVL-NEXT:    [[TMP23:%.*]] = mul i64 -1, [[TMP30]]
 ; IF-EVL-NEXT:    [[TMP24:%.*]] = getelementptr i32, ptr [[TMP21]], i64 [[TMP22]]
 ; IF-EVL-NEXT:    [[TMP25:%.*]] = getelementptr i32, ptr [[TMP24]], i64 [[TMP23]]
-; IF-EVL-NEXT:    [[VP_REVERSE5:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_REVERSE]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[VP_REVERSE_MASK6:%.*]] = call <vscale x 4 x i1> @llvm.experimental.vp.reverse.nxv4i1(<vscale x 4 x i1> [[TMP14]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
 ; IF-EVL-NEXT:    call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_REVERSE5]], ptr align 4 [[TMP25]], <vscale x 4 x i1> [[VP_REVERSE_MASK6]], i32 [[TMP5]])
 ; IF-EVL-NEXT:    [[TMP28:%.*]] = zext i32 [[TMP5]] to i64
@@ -240,8 +240,8 @@ define void @reverse_load_store_masked(i64 %startval, ptr noalias %ptr, ptr noal
 ; NO-VP-NEXT:    [[TMP20:%.*]] = mul i64 -1, [[TMP19]]
 ; NO-VP-NEXT:    [[TMP21:%.*]] = getelementptr i32, ptr [[TMP17]], i64 [[TMP18]]
 ; NO-VP-NEXT:    [[TMP22:%.*]] = getelementptr i32, ptr [[TMP21]], i64 [[TMP20]]
-; NO-VP-NEXT:    [[REVERSE3:%.*]] = call <vscale x 4 x i1> @llvm.vector.reverse.nxv4i1(<vscale x 4 x i1> [[TMP10]])
 ; NO-VP-NEXT:    [[REVERSE4:%.*]] = call <vscale x 4 x i32> @llvm.vector.reverse.nxv4i32(<vscale x 4 x i32> [[REVERSE2]])
+; NO-VP-NEXT:    [[REVERSE3:%.*]] = call <vscale x 4 x i1> @llvm.vector.reverse.nxv4i1(<vscale x 4 x i1> [[TMP10]])
 ; NO-VP-NEXT:    call void @llvm.masked.store.nxv4i32.p0(<vscale x 4 x i32> [[REVERSE4]], ptr [[TMP22]], i32 4, <vscale x 4 x i1> [[REVERSE3]])
 ; NO-VP-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]
 ; NO-VP-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -328,22 +328,22 @@ define void @multiple_reverse_vector_pointer(ptr noalias %a, ptr noalias %b, ptr
 ; IF-EVL-NEXT:    [[TMP14:%.*]] = getelementptr i8, ptr [[B:%.*]], <vscale x 16 x i8> [[VP_REVERSE]]
 ; IF-EVL-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 16 x i8> @llvm.vp.gather.nxv16i8.nxv16p0(<vscale x 16 x ptr> align 1 [[TMP14]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
 ; IF-EVL-NEXT:    [[TMP15:%.*]] = getelementptr i8, ptr [[C:%.*]], i64 [[OFFSET_IDX]]
+; IF-EVL-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vp.reverse.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
 ; IF-EVL-NEXT:    [[TMP16:%.*]] = zext i32 [[TMP6]] to i64
 ; IF-EVL-NEXT:    [[TMP17:%.*]] = mul i64 0, [[TMP16]]
 ; IF-EVL-NEXT:    [[TMP30:%.*]] = sub i64 [[TMP16]], 1
 ; IF-EVL-NEXT:    [[TMP18:%.*]] = mul i64 -1, [[TMP30]]
 ; IF-EVL-NEXT:    [[TMP19:%.*]] = getelementptr i8, ptr [[TMP15]], i64 [[TMP17]]
 ; IF-EVL-NEXT:    [[TMP20:%.*]] = getelementptr i8, ptr [[TMP19]], i64 [[TMP18]]
-; IF-EVL-NEXT:    [[VP_REVERSE1:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vp.reverse.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
 ; IF-EVL-NEXT:    call void @llvm.vp.store.nxv16i8.p0(<vscale x 16 x i8> [[VP_REVERSE1]], ptr align 1 [[TMP20]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
 ; IF-EVL-NEXT:    [[TMP21:%.*]] = getelementptr i8, ptr [[D:%.*]], i64 [[OFFSET_IDX]]
+; IF-EVL-NEXT:    [[VP_REVERSE2:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vp.reverse.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
 ; IF-EVL-NEXT:    [[TMP22:%.*]] = zext i32 [[TMP6]] to i64
 ; IF-EVL-NEXT:    [[TMP23:%.*]] = mul i64 0, [[TMP22]]
 ; IF-EVL-NEXT:    [[TMP31:%.*]] = sub i64 [[TMP22]], 1
 ; IF-EVL-NEXT:    [[TMP24:%.*]] = mul i64 -1, [[TMP31]]
 ; IF-EVL-NEXT:    [[TMP25:%.*]] = getelementptr i8, ptr [[TMP21]], i64 [[TMP23]]
 ; IF-EVL-NEXT:    [[TMP26:%.*]] = getelementptr i8, ptr [[TMP25]], i64 [[TMP24]]
-; IF-EVL-NEXT:    [[VP_REVERSE2:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vp.reverse.nxv16i8(<vscale x 16 x i8> [[WIDE_MASKED_GATHER]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
 ; IF-EVL-NEXT:    call void @llvm.vp.store.nxv16i8.p0(<vscale x 16 x i8> [[VP_REVERSE2]], ptr align 1 [[TMP26]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
 ; IF-EVL-NEXT:    [[TMP27:%.*]] = zext i32 [[TMP6]] to i64
 ; IF-EVL-NEXT:    [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP27]], [[EVL_BASED_IV]]
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-uniform-store.ll b/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-uniform-store.ll
index aff610e6bec22..9553048e0b6d3 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-uniform-store.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-uniform-store.ll
@@ -15,37 +15,37 @@ define void @lshift_significand(i32 %n, ptr nocapture writeonly %dst) {
 ; CHECK-NEXT:    [[TMP0:%.*]] = sub i64 3, [[SPEC_SELECT]]
 ; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
 ; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[REVERSE:%.*]] = call <vscale x 2 x i64> @llvm.vector.reverse.nxv2i64(<vscale x 2 x i64> zeroinitializer)
 ; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
 ; CHECK:       [[VECTOR_BODY]]:
 ; CHECK-NEXT:    [[EVL_BASED_IV:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_EVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[AVL:%.*]] = phi i64 [ [[TMP0]], %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[TMP11:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 2, i1 true)
+; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 2, i1 true)
 ; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i64 [[SPEC_SELECT]], [[EVL_BASED_IV]]
 ; CHECK-NEXT:    [[TMP12:%.*]] = sub nuw nsw i64 1, [[OFFSET_IDX]]
 ; CHECK-NEXT:    [[ARRAYIDX13:%.*]] = getelementptr i64, ptr [[DST]], i64 [[TMP12]]
-; CHECK-NEXT:    [[TMP15:%.*]] = zext i32 [[TMP11]] to i64
-; CHECK-NEXT:    [[TMP16:%.*]] = mul i64 0, [[TMP15]]
-; CHECK-NEXT:    [[TMP23:%.*]] = sub i64 [[TMP15]], 1
-; CHECK-NEXT:    [[TMP17:%.*]] = mul i64 -1, [[TMP23]]
-; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr i64, ptr [[ARRAYIDX13]], i64 [[TMP16]]
-; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr i64, ptr [[TMP18]], i64 [[TMP17]]
-; CHECK-NEXT:    [[VP_REVERSE:%.*]] = call <vscale x 2 x i64> @llvm.experimental.vp.reverse.nxv2i64(<vscale x 2 x i64> zeroinitializer, <vscale x 2 x i1> splat (i1 true), i32 [[TMP11]])
-; CHECK-NEXT:    call void @llvm.vp.store.nxv2i64.p0(<vscale x 2 x i64> [[VP_REVERSE]], ptr align 8 [[TMP19]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP11]])
-; CHECK-NEXT:    [[TMP20:%.*]] = zext i32 [[TMP11]] to i64
-; CHECK-NEXT:    [[INDEX_EVL_NEXT]] = add i64 [[TMP20]], [[EVL_BASED_IV]]
-; CHECK-NEXT:    [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP20]]
-; CHECK-NEXT:    [[TMP21:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
-; CHECK-NEXT:    br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT:    [[TMP4:%.*]] = zext i32 [[TMP1]] to i64
+; CHECK-NEXT:    [[TMP5:%.*]] = mul i64 0, [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = sub i64 [[TMP4]], 1
+; CHECK-NEXT:    [[TMP7:%.*]] = mul i64 -1, [[TMP6]]
+; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr i64, ptr [[ARRAYIDX13]], i64 [[TMP5]]
+; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr i64, ptr [[TMP8]], i64 [[TMP7]]
+; CHECK-NEXT:    call void @llvm.vp.store.nxv2i64.p0(<vscale x 2 x i64> [[REVERSE]], ptr align 8 [[TMP9]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP1]])
+; CHECK-NEXT:    [[TMP10:%.*]] = zext i32 [[TMP1]] to i64
+; CHECK-NEXT:    [[INDEX_EVL_NEXT]] = add i64 [[TMP10]], [[EVL_BASED_IV]]
+; CHECK-NEXT:    [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP10]]
+; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
+; CHECK-NEXT:    br i1 [[TMP11]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    br label %[[EXIT:.*]]
 ; CHECK:       [[SCALAR_PH]]:
 ; CHECK-NEXT:    br label %[[LOOP:.*]]
 ; CHECK:       [[LOOP]]:
-; CHECK-NEXT:    [[IV1:%.*]] = phi i64 [ [[SPEC_SELECT]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
-; CHECK-NEXT:    [[TMP22:%.*]] = sub nuw nsw i64 1, [[IV1]]
-; CHECK-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr i64, ptr [[DST]], i64 [[TMP22]]
+; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[SPEC_SELECT]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[TMP13:%.*]] = sub nuw nsw i64 1, [[IV]]
+; CHECK-NEXT:    [[ARRAYIDX14:%.*]] = getelementptr i64, ptr [[DST]], i64 [[TMP13]]
 ; CHECK-NEXT:    store i64 0, ptr [[ARRAYIDX14]], align 8
-; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV1]], 1
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
 ; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 3
 ; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP4:![0-9]+]]
 ; CHECK:       [[EXIT]]:
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/vplan-riscv-vector-reverse.ll b/llvm/test/Transforms/LoopVectorize/RISCV/vplan-riscv-vector-reverse.ll
index 46695221c27db..8b6ce538474c7 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/vplan-riscv-vector-reverse.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/vplan-riscv-vector-reverse.ll
@@ -34,11 +34,13 @@ define void @vector_reverse_i64(ptr nocapture noundef writeonly %A, ptr nocaptur
 ; CHECK-NEXT:     CLONE ir<[[IDX_PROM:%.+]]> = zext ir<[[IDX]]>
 ; CHECK-NEXT:     CLONE ir<[[ARRAY_IDX_B:%.+]]> = getelementptr inbounds ir<[[B:%.+]]>, ir<[[IDX_PROM]]>
 ; CHECK-NEXT:     vp<[[VEC_END_PTR_B:%.+]]> = vector-end-pointer ir<[[ARRAY_IDX_B]]>, vp<[[EVL]]>
-; CHECK-NEXT:     WIDEN ir<[[VAL_B:%.+]]> = vp.load vp<[[VEC_END_PTR_B]]>, vp<[[EVL]]>
-; CHECK-NEXT:     WIDEN ir<[[ADD_RESULT:%.+]]> = add ir<[[VAL_B]]>, ir<1>
+; CHECK-NEXT:     WIDEN ir<[[LOAD_B:%.+]]> = vp.load vp<[[VEC_END_PTR_B]]>, vp<[[EVL]]>
+; CHECK-NEXT:     WIDEN-INTRINSIC vp<[[VAL_B:%.+]]> = call llvm.experimental.vp.reverse(ir<[[LOAD_B]]>, ir<true>, vp<[[EVL]]>)
+; CHECK-NEXT:     WIDEN ir<[[ADD_RESULT:%.+]]> = add vp<[[VAL_B]]>, ir<1>
 ; CHECK-NEXT:     CLONE ir<[[ARRAY_IDX_A:%.+]]> = getelementptr inbounds ir<[[A:%.+]]>, ir<[[IDX_PROM]]>
+; CHECK-NEXT:     WIDEN-INTRINSIC vp<[[STORE_VAL:%.+]]> = call llvm.experimental.vp.reverse(ir<[[ADD_RESULT]]>, ir<true>, vp<[[EVL]]>)
 ; CHECK-NEXT:     vp<[[VEC_END_PTR_A:%.+]]> = vector-end-pointer ir<[[ARRAY_IDX_A]]>, vp<[[EVL]]>
-; CHECK-NEXT:     WIDEN vp.store vp<[[VEC_END_PTR_A]]>, ir<[[ADD_RESULT]]>, vp<[[EVL]]>
+; CHECK-NEXT:     WIDEN vp.store vp<[[VEC_END_PTR_A]]>, vp<[[STORE_VAL]]>, vp<[[EVL]]>
 ; CHECK-NEXT:     EMIT vp<[[IV_NEXT]]> = add vp<[[EVL]]>, vp<[[EVL_PHI]]>
 ; CHECK-NEXT:     EMIT vp<[[AVL_NEXT]]> = sub nuw vp<[[AVL]]>, vp<[[EVL]]>
 ; CHECK-NEXT:     EMIT vp<[[INDEX_NEXT]]> = add vp<[[INDUCTION]]>, vp<[[VFxUF]]>
diff --git a/llvm/test/Transforms/LoopVectorize/X86/masked_load_store.ll b/llvm/test/Transforms/LoopVectorize/X86/masked_load_store.ll
index 5065dc89014f9..de69b6a39d46c 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/masked_load_store.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/masked_load_store.ll
@@ -1391,12 +1391,12 @@ define void @foo6(ptr nocapture readonly %in, ptr nocapture %out, i32 %size, ptr
 ; AVX2-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 -12
 ; AVX2-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[TMP8]], i32 -3
 ; AVX2-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP3]], align 4, !alias.scope [[META21:![0-9]+]]
-; AVX2-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x i32>, ptr [[TMP5]], align 4, !alias.scope [[META21]]
-; AVX2-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD6]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x i32>, ptr [[TMP7]], align 4, !alias.scope [[META21]]
-; AVX2-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD8]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:    [[WIDE_LOAD10:%.*]] = load <4 x i32>, ptr [[TMP9]], align 4, !alias.scope [[META21]]
+; AVX2-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD6]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD8]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:    [[REVERSE11:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD10]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:    [[TMP10:%.*]] = icmp sgt <4 x i32> [[REVERSE]], zeroinitializer
 ; AVX2-NEXT:    [[TMP11:%.*]] = icmp sgt <4 x i32> [[REVERSE7]], zeroinitializer
@@ -1412,21 +1412,21 @@ define void @foo6(ptr nocapture readonly %in, ptr nocapture %out, i32 %size, ptr
 ; AVX2-NEXT:    [[TMP21:%.*]] = getelementptr double, ptr [[TMP14]], i32 -12
 ; AVX2-NEXT:    [[TMP22:%.*]] = getelementptr double, ptr [[TMP21]], i32 -3
 ; AVX2-NEXT:    [[REVERSE12:%.*]] = shufflevector <4 x i1> [[TMP10]], <4 x i1> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0(ptr [[TMP16]], i32 8, <4 x i1> [[REVERSE12]], <4 x double> poison), !alias.scope [[META24:![0-9]+]]
-; AVX2-NEXT:    [[REVERSE13:%.*]] = shufflevector <4 x double> [[WIDE_MASKED_LOAD]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    [[WIDE_MASKED_LOAD21:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0(ptr [[TMP16]], i32 8, <4 x i1> [[REVERSE12]], <4 x double> poison), !alias.scope [[META24:![0-9]+]]
 ; AVX2-NEXT:    [[REVERSE14:%.*]] = shufflevector <4 x i1> [[TMP11]], <4 x i1> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:    [[WIDE_MASKED_LOAD15:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0(ptr [[TMP18]], i32 8, <4 x i1> [[REVERSE14]], <4 x double> poison), !alias.scope [[META24]]
-; AVX2-NEXT:    [[REVERSE16:%.*]] = shufflevector <4 x double> [[WIDE_MASKED_LOAD15]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    [[WIDE_MASKED_LOAD14:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0(ptr [[TMP18]], i32 8, <4 x i1> [[REVERSE14]], <4 x double> poison), !alias.scope [[META24]]
 ; AVX2-NEXT:    [[REVERSE17:%.*]] = shufflevector <4 x i1> [[TMP12]], <4 x i1> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:    [[WIDE_MASKED_LOAD18:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0(ptr [[TMP20]], i32 8, <4 x i1> [[REVERSE17]], <4 x double> poison), !alias.scope [[META24]]
-; AVX2-NEXT:    [[REVERSE19:%.*]] = shufflevector <4 x double> [[WIDE_MASKED_LOAD18]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    [[WIDE_MASKED_LOAD16:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0(ptr [[TMP20]], i32 8, <4 x i1> [[REVERSE17]], <4 x double> poison), !alias.scope [[META24]]
 ; AVX2-NEXT:    [[REVERSE20:%.*]] = shufflevector <4 x i1> [[TMP13]], <4 x i1> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:    [[WIDE_MASKED_LOAD21:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0(ptr [[TMP22]], i32 8, <4 x i1> [[REVERSE20]], <4 x double> poison), !alias.scope [[META24]]
+; AVX2-NEXT:    [[WIDE_MASKED_LOAD18:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0(ptr [[TMP22]], i32 8, <4 x i1> [[REVERSE20]], <4 x double> poison), !alias.scope [[META24]]
 ; AVX2-NEXT:    [[REVERSE22:%.*]] = shufflevector <4 x double> [[WIDE_MASKED_LOAD21]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:    [[TMP23:%.*]] = fadd <4 x double> [[REVERSE13]], splat (double 5.000000e-01)
+; AVX2-NEXT:    [[REVERSE16:%.*]] = shufflevector <4 x double> [[WIDE_MASKED_LOAD14]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    [[REVERSE19:%.*]] = shufflevector <4 x double> [[WIDE_MASKED_LOAD16]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    [[REVERSE23:%.*]] = shufflevector <4 x double> [[WIDE_MASKED_LOAD18]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    [[TMP23:%.*]] = fadd <4 x double> [[REVERSE22]], splat (double 5.000000e-01)
 ; AVX2-NEXT:    [[TMP24:%.*]] = fadd <4 x double> [[REVERSE16]], splat (double 5.000000e-01)
 ; AVX2-NEXT:    [[TMP25:%.*]] = fadd <4 x double> [[REVERSE19]], splat (double 5.000000e-01)
-; AVX2-NEXT:    [[TMP26:%.*]] = fadd <4 x double> [[REVERSE22]], splat (double 5.000000e-01)
+; AVX2-NEXT:    [[TMP26:%.*]] = fadd <4 x double> [[REVERSE23]], splat (double 5.000000e-01)
 ; AVX2-NEXT:    [[TMP27:%.*]] = getelementptr double, ptr [[OUT]], i64 [[OFFSET_IDX]]
 ; AVX2-NEXT:    [[TMP28:%.*]] = getelementptr double, ptr [[TMP27]], i32 0
 ; AVX2-NEXT:    [[TMP29:%.*]] = getelementptr double, ptr [[TMP28]], i32 -3
@@ -1437,12 +1437,12 @@ define void @foo6(ptr nocapture readonly %in, ptr nocapture %out, i32 %size, ptr
 ; AVX2-NEXT:    [[TMP34:%.*]] = getelementptr double, ptr [[TMP27]], i32 -12
 ; AVX2-NEXT:    [[TMP35:%.*]] = getelementptr double, ptr [[TMP34]], i32 -3
 ; AVX2-NEXT:    [[REVERSE24:%.*]] = shufflevector <4 x double> [[TMP23]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:    call void @llvm.masked.store.v4f64.p0(<4 x double> [[REVERSE24]], ptr [[TMP29]], i32 8, <4 x i1> [[REVERSE12]]), !alias.scope [[META26:![0-9]+]], !noalias [[META28:![0-9]+]]
 ; AVX2-NEXT:    [[REVERSE26:%.*]] = shufflevector <4 x double> [[TMP24]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:    call void @llvm.masked.store.v4f64.p0(<4 x double> [[REVERSE26]], ptr [[TMP31]], i32 8, <4 x i1> [[REVERSE14]]), !alias.scope [[META26]], !noalias [[META28]]
 ; AVX2-NEXT:    [[REVERSE28:%.*]] = shufflevector <4 x double> [[TMP25]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:    call void @llvm.masked.store.v4f64.p0(<4 x double> [[REVERSE28]], ptr [[TMP33]], i32 8, <4 x i1> [[REVERSE17]]), !alias.scope [[META26]], !noalias [[META28]]
 ; AVX2-NEXT:    [[REVERSE30:%.*]] = shufflevector <4 x double> [[TMP26]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:    call void @llvm.masked.store.v4f64.p0(<4 x double> [[REVERSE24]], ptr [[TMP29]], i32 8, <4 x i1> [[REVERSE12]]), !alias.scope [[META26:![0-9]+]], !noalias [[META28:![0-9]+]]
+; AVX2-NEXT:    call void @llvm.masked.store.v4f64.p0(<4 x double> [[REVERSE26]], ptr [[TMP31]], i32 8, <4 x i1> [[REVERSE14]]), !alias.scope [[META26]], !noalias [[META28]]
+; AVX2-NEXT:    call void @llvm.masked.store.v4f64.p0(<4 x double> [[REVERSE28]], ptr [[TMP33]], i32 8, <4 x i1> [[REVERSE17]]), !alias.scope [[META26]], !noalias [[META28]]
 ; AVX2-NEXT:    call void @llvm.masked.store.v4f64.p0(<4 x double> [[REVERSE30]], ptr [[TMP35]], i32 8, <4 x i1> [[REVERSE20]]), !alias.scope [[META26]], !noalias [[META28]]
 ; AVX2-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
 ; AVX2-NEXT:    [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096
@@ -1502,13 +1502,13 @@ define void @foo6(ptr nocapture readonly %in, ptr nocapture %out, i32 %size, ptr
 ; AVX512-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 -24
 ; AVX512-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[TMP8]], i32 -7
 ; AVX512-NEXT:    [[WIDE_LOAD:%.*]] = load <8 x i32>, ptr [[TMP3]], align 4, !alias.scope [[META32:![0-9]+]]
-; AVX512-NEXT:    [[REVERSE:%.*]] = shufflevector <8 x i32> [[WIDE_LOAD]], <8 x i32> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; AVX512-NEXT:    [[WIDE_LOAD6:%.*]] = load <8 x i32>, ptr [[TMP5]], align 4, !alias.scope [[META32]]
+; AVX512-NEXT:    [[WIDE_LOAD7:%.*]] = load <8 x i32>, ptr [[TMP7]], align 4, !alias.scope [[META32]]
+; AVX512-NEXT:    [[WIDE_LOAD8:%.*]] = load <8 x i32>, ptr [[TMP9]], align 4, !alias.scope [[META32]]
+; AVX512-NEXT:    [[REVERSE:%.*]] = shufflevector <8 x i32> [[WIDE_LOAD]], <8 x i32> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; AVX512-NEXT:    [[REVERSE7:%.*]] = shufflevector <8 x i32> [[WIDE_LOAD6]], <8 x i32> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[WIDE_LOAD8:%.*]] = load <8 x i32>, ptr [[TMP7]], align 4, !alias.scope [[META32]]
-; AVX512-NEXT:    [[REVERSE9:%.*]] = shufflevector <8 x i32> [[WIDE_LOAD8]], <8 x i32> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[WIDE_LOAD10:%.*]] = load <8 x i32>, ptr [[TMP9]], align 4, !alias.scope [[META32]]
-; AVX512-NEXT:    [[REVERSE11:%.*]] = shufflevector <8 x i32> [[WIDE_LOAD10]], <8 x i32> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[REVERSE9:%.*]] = shufflevector <8 x i32> [[WIDE_LOAD7]], <8 x i32> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[REVERSE11:%.*]] = shufflevector <8 x i32> [[WIDE_LOAD8]], <8 x i32> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; AVX512-NEXT:    [[TMP10:%.*]] = icmp sgt <8 x i32> [[REVERSE]], zeroinitializer
 ; AVX512-NEXT:    [[TMP11:%.*]] = icmp sgt <8 x i32> [[REVERSE7]], zeroinitializer
 ; AVX512-NEXT:    [[TMP12:%.*]] = icmp sgt <8 x i32> [[REVERSE9]], zeroinitializer
@@ -1524,20 +1524,20 @@ define void @foo6(ptr nocapture readonly %in, ptr nocapture %out, i32 %size, ptr
 ; AVX512-NEXT:    [[TMP22:%.*]] = getelementptr double, ptr [[TMP21]], i32 -7
 ; AVX512-NEXT:    [[REVERSE12:%.*]] = shufflevector <8 x i1> [[TMP10]], <8 x i1> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; AVX512-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <8 x double> @llvm.masked.load.v8f64.p0(ptr [[TMP16]], i32 8, <8 x i1> [[REVERSE12]], <8 x double> poison), !alias.scope [[META35:![0-9]+]]
-; AVX512-NEXT:    [[REVERSE13:%.*]] = shufflevector <8 x double> [[WIDE_MASKED_LOAD]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[REVERSE14:%.*]] = shufflevector <8 x i1> [[TMP11]], <8 x i1> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[WIDE_MASKED_LOAD15:%.*]] = call <8 x double> @llvm.masked.load.v8f64.p0(ptr [[TMP18]], i32 8, <8 x i1> [[REVERSE14]], <8 x double> poison), !alias.scope [[META35]]
-; AVX512-NEXT:    [[REVERSE16:%.*]] = shufflevector <8 x double> [[WIDE_MASKED_LOAD15]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[REVERSE17:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[WIDE_MASKED_LOAD18:%.*]] = call <8 x double> @llvm.masked.load.v8f64.p0(ptr [[TMP20]], i32 8, <8 x i1> [[REVERSE17]], <8 x double> poison), !alias.scope [[META35]]
-; AVX512-NEXT:    [[REVERSE19:%.*]] = shufflevector <8 x double> [[WIDE_MASKED_LOAD18]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[REVERSE20:%.*]] = shufflevector <8 x i1> [[TMP13]], <8 x i1> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[WIDE_MASKED_LOAD21:%.*]] = call <8 x double> @llvm.masked.load.v8f64.p0(ptr [[TMP22]], i32 8, <8 x i1> [[REVERSE20]], <8 x double> poison), !alias.scope [[META35]]
-; AVX512-NEXT:    [[REVERSE22:%.*]] = shufflevector <8 x double> [[WIDE_MASKED_LOAD21]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    [[TMP23:%.*]] = fadd <8 x double> [[REVERSE13]], splat (double 5.000000e-01)
+; AVX512-NEXT:    [[REVERSE13:%.*]] = shufflevector <8 x i1> [[TMP11]], <8 x i1> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[WIDE_MASKED_LOAD14:%.*]] = call <8 x double> @llvm.masked.load.v8f64.p0(ptr [[TMP18]], i32 8, <8 x i1> [[REVERSE13]], <8 x double> poison), !alias.scope [[META35]]
+; AVX512-NEXT:    [[REVERSE15:%.*]] = shufflevector <8 x i1> [[TMP12]], <8 x i1> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[WIDE_MASKED_LOAD16:%.*]] = call <8 x double> @llvm.masked.load.v8f64.p0(ptr [[TMP20]], i32 8, <8 x i1> [[REVERSE15]], <8 x double> poison), !alias.scope [[META35]]
+; AVX512-NEXT:    [[REVERSE17:%.*]] = shufflevector <8 x i1> [[TMP13]], <8 x i1> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[WIDE_MASKED_LOAD18:%.*]] = call <8 x double> @llvm.masked.load.v8f64.p0(ptr [[TMP22]], i32 8, <8 x i1> [[REVERSE17]], <8 x double> poison), !alias.scope [[META35]]
+; AVX512-NEXT:    [[REVERSE22:%.*]] = shufflevector <8 x double> [[WIDE_MASKED_LOAD]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[REVERSE16:%.*]] = shufflevector <8 x double> [[WIDE_MASKED_LOAD14]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[REVERSE19:%.*]] = shufflevector <8 x double> [[WIDE_MASKED_LOAD16]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[REVERSE23:%.*]] = shufflevector <8 x double> [[WIDE_MASKED_LOAD18]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[TMP23:%.*]] = fadd <8 x double> [[REVERSE22]], splat (double 5.000000e-01)
 ; AVX512-NEXT:    [[TMP24:%.*]] = fadd <8 x double> [[REVERSE16]], splat (double 5.000000e-01)
 ; AVX512-NEXT:    [[TMP25:%.*]] = fadd <8 x double> [[REVERSE19]], splat (double 5.000000e-01)
-; AVX512-NEXT:    [[TMP26:%.*]] = fadd <8 x double> [[REVERSE22]], splat (double 5.000000e-01)
+; AVX512-NEXT:    [[TMP26:%.*]] = fadd <8 x double> [[REVERSE23]], splat (double 5.000000e-01)
 ; AVX512-NEXT:    [[TMP27:%.*]] = getelementptr double, ptr [[OUT]], i64 [[OFFSET_IDX]]
 ; AVX512-NEXT:    [[TMP28:%.*]] = getelementptr double, ptr [[TMP27]], i32 0
 ; AVX512-NEXT:    [[TMP29:%.*]] = getelementptr double, ptr [[TMP28]], i32 -7
@@ -1548,13 +1548,13 @@ define void @foo6(ptr nocapture readonly %in, ptr nocapture %out, i32 %size, ptr
 ; AVX512-NEXT:    [[TMP34:%.*]] = getelementptr double, ptr [[TMP27]], i32 -24
 ; AVX512-NEXT:    [[TMP35:%.*]] = getelementptr double, ptr [[TMP34]], i32 -7
 ; AVX512-NEXT:    [[REVERSE24:%.*]] = shufflevector <8 x double> [[TMP23]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[REVERSE27:%.*]] = shufflevector <8 x double> [[TMP24]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[REVERSE25:%.*]] = shufflevector <8 x double> [[TMP25]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:    [[REVERSE26:%.*]] = shufflevector <8 x double> [[TMP26]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; AVX512-NEXT:    call void @llvm.masked.store.v8f64.p0(<8 x double> [[REVERSE24]], ptr [[TMP29]], i32 8, <8 x i1> [[REVERSE12]]), !alias.scope [[META37:![0-9]+]], !noalias [[META39:![0-9]+]]
-; AVX512-NEXT:    [[REVERSE26:%.*]] = shufflevector <8 x double> [[TMP24]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    call void @llvm.masked.store.v8f64.p0(<8 x double> [[REVERSE26]], ptr [[TMP31]], i32 8, <8 x i1> [[REVERSE14]]), !alias.scope [[META37]], !noalias [[META39]]
-; AVX512-NEXT:    [[REVERSE28:%.*]] = shufflevector <8 x double> [[TMP25]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    call void @llvm.masked.store.v8f64.p0(<8 x double> [[REVERSE28]], ptr [[TMP33]], i32 8, <8 x i1> [[REVERSE17]]), !alias.scope [[META37]], !noalias [[META39]]
-; AVX512-NEXT:    [[REVERSE30:%.*]] = shufflevector <8 x double> [[TMP26]], <8 x double> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:    call void @llvm.masked.store.v8f64.p0(<8 x double> [[REVERSE30]], ptr [[TMP35]], i32 8, <8 x i1> [[REVERSE20]]), !alias.scope [[META37]], !noalias [[META39]]
+; AVX512-NEXT:    call void @llvm.masked.store.v8f64.p0(<8 x double> [[REVERSE27]], ptr [[TMP31]], i32 8, <8 x i1> [[REVERSE13]]), !alias.scope [[META37]], !noalias [[META39]]
+; AVX512-NEXT:    call void @llvm.masked.store.v8f64.p0(<8 x double> [[REVERSE25]], ptr [[TMP33]], i32 8, <8 x i1> [[REVERSE15]]), !alias.scope [[META37]], !noalias [[META39]]
+; AVX512-NEXT:    call void @llvm.masked.store.v8f64.p0(<8 x double> [[REVERSE26]], ptr [[TMP35]], i32 8, <8 x i1> [[REVERSE17]]), !alias.scope [[META37]], !noalias [[META39]]
 ; AVX512-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
 ; AVX512-NEXT:    [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096
 ; AVX512-NEXT:    br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP40:![0-9]+]]
diff --git a/llvm/test/Transforms/LoopVectorize/interleave-with-i65-induction.ll b/llvm/test/Transforms/LoopVectorize/interleave-with-i65-induction.ll
index 3eb16e9a2d78e..aa631573cbead 100644
--- a/llvm/test/Transforms/LoopVectorize/interleave-with-i65-induction.ll
+++ b/llvm/test/Transforms/LoopVectorize/interleave-with-i65-induction.ll
@@ -24,8 +24,8 @@ define void @i65_induction_with_negative_step(ptr %dst) {
 ; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[TMP4]], i32 -4
 ; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[TMP7]], i32 -3
 ; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i64> [[TMP2]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    store <4 x i64> [[REVERSE]], ptr [[TMP6]], align 8
 ; CHECK-NEXT:    [[REVERSE1:%.*]] = shufflevector <4 x i64> [[TMP3]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    store <4 x i64> [[REVERSE]], ptr [[TMP6]], align 8
 ; CHECK-NEXT:    store <4 x i64> [[REVERSE1]], ptr [[TMP8]], align 8
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i65 [[INDEX]], 8
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[STEP_ADD]], splat (i64 -4)
diff --git a/llvm/test/Transforms/LoopVectorize/iv-select-cmp-decreasing.ll b/llvm/test/Transforms/LoopVectorize/iv-select-cmp-decreasing.ll
index 615f50124b41d..c26a925533c7c 100644
--- a/llvm/test/Transforms/LoopVectorize/iv-select-cmp-decreasing.ll
+++ b/llvm/test/Transforms/LoopVectorize/iv-select-cmp-decreasing.ll
@@ -74,12 +74,12 @@ define i64 @select_decreasing_induction_icmp_const_start(ptr %a) {
 ; IC4VF4-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i32 -12
 ; IC4VF4-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[TMP7]], i32 -3
 ; IC4VF4-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP2]], align 8
-; IC4VF4-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP4]], align 8
-; IC4VF4-NEXT:    [[REVERSE5:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD4]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x i64>, ptr [[TMP6]], align 8
-; IC4VF4-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD6]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x i64>, ptr [[TMP8]], align 8
+; IC4VF4-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; IC4VF4-NEXT:    [[REVERSE5:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD4]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; IC4VF4-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD6]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD8]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[TMP9:%.*]] = icmp sgt <4 x i64> [[REVERSE]], splat (i64 3)
 ; IC4VF4-NEXT:    [[TMP10:%.*]] = icmp sgt <4 x i64> [[REVERSE5]], splat (i64 3)
@@ -997,12 +997,12 @@ define i64 @select_decreasing_induction_icmp_iv_unsigned(ptr %a) {
 ; IC4VF4-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[TMP0]], i32 -12
 ; IC4VF4-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[TMP7]], i32 -3
 ; IC4VF4-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP2]], align 8
-; IC4VF4-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP4]], align 8
-; IC4VF4-NEXT:    [[REVERSE5:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD4]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x i64>, ptr [[TMP6]], align 8
-; IC4VF4-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD6]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x i64>, ptr [[TMP8]], align 8
+; IC4VF4-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; IC4VF4-NEXT:    [[REVERSE5:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD4]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; IC4VF4-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD6]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x i64> [[WIDE_LOAD8]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; IC4VF4-NEXT:    [[TMP9:%.*]] = icmp sgt <4 x i64> [[REVERSE]], splat (i64 3)
 ; IC4VF4-NEXT:    [[TMP10:%.*]] = icmp sgt <4 x i64> [[REVERSE5]], splat (i64 3)
diff --git a/llvm/test/Transforms/LoopVectorize/reverse_induction.ll b/llvm/test/Transforms/LoopVectorize/reverse_induction.ll
index 4895f6a83d56d..55eb1dd2109a9 100644
--- a/llvm/test/Transforms/LoopVectorize/reverse_induction.ll
+++ b/llvm/test/Transforms/LoopVectorize/reverse_induction.ll
@@ -25,8 +25,8 @@ define i32 @reverse_induction_i64(i64 %startval, ptr %ptr) {
 ; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[TMP5]], i32 -4
 ; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[TMP8]], i32 -3
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP7]], align 4
-; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[WIDE_LOAD3:%.*]] = load <4 x i32>, ptr [[TMP9]], align 4
+; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[REVERSE4:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD3]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[TMP10]] = add <4 x i32> [[REVERSE]], [[VEC_PHI]]
 ; CHECK-NEXT:    [[TMP11]] = add <4 x i32> [[REVERSE4]], [[VEC_PHI2]]
@@ -93,8 +93,8 @@ define i32 @reverse_induction_i128(i128 %startval, ptr %ptr) {
 ; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[TMP5]], i32 -4
 ; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[TMP8]], i32 -3
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP7]], align 4
-; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[WIDE_LOAD3:%.*]] = load <4 x i32>, ptr [[TMP9]], align 4
+; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[REVERSE4:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD3]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[TMP10]] = add <4 x i32> [[REVERSE]], [[VEC_PHI]]
 ; CHECK-NEXT:    [[TMP11]] = add <4 x i32> [[REVERSE4]], [[VEC_PHI2]]
@@ -171,8 +171,8 @@ define i32 @reverse_induction_i16(i16 %startval, ptr %ptr) {
 ; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, ptr [[TMP9]], i32 -4
 ; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[TMP12]], i32 -3
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP11]], align 4
-; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[WIDE_LOAD3:%.*]] = load <4 x i32>, ptr [[TMP13]], align 4
+; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[REVERSE4:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD3]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:    [[TMP14]] = add <4 x i32> [[REVERSE]], [[VEC_PHI]]
 ; CHECK-NEXT:    [[TMP15]] = add <4 x i32> [[REVERSE4]], [[VEC_PHI2]]
@@ -261,8 +261,8 @@ define void @reverse_forward_induction_i64_i8() {
 ; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i32 -4
 ; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[TMP10]], i32 -3
 ; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[TMP4]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    store <4 x i32> [[REVERSE]], ptr [[TMP9]], align 4
 ; CHECK-NEXT:    [[REVERSE2:%.*]] = shufflevector <4 x i32> [[TMP5]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    store <4 x i32> [[REVERSE]], ptr [[TMP9]], align 4
 ; CHECK-NEXT:    store <4 x i32> [[REVERSE2]], ptr [[TMP11]], align 4
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i8> [[STEP_ADD]], splat (i8 4)
@@ -327,8 +327,8 @@ define void @reverse_forward_induction_i64_i8_signed() {
 ; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i32 -4
 ; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[TMP10]], i32 -3
 ; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[TMP4]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    store <4 x i32> [[REVERSE]], ptr [[TMP9]], align 4
 ; CHECK-NEXT:    [[REVERSE2:%.*]] = shufflevector <4 x i32> [[TMP5]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    store <4 x i32> [[REVERSE]], ptr [[TMP9]], align 4
 ; CHECK-NEXT:    store <4 x i32> [[REVERSE2]], ptr [[TMP11]], align 4
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i8> [[STEP_ADD]], splat (i8 4)
diff --git a/llvm/test/Transforms/LoopVectorize/single-early-exit-interleave.ll b/llvm/test/Transforms/LoopVectorize/single-early-exit-interleave.ll
index 4d913b19129d9..f8e4207f65ef9 100644
--- a/llvm/test/Transforms/LoopVectorize/single-early-exit-interleave.ll
+++ b/llvm/test/Transforms/LoopVectorize/single-early-exit-interleave.ll
@@ -663,12 +663,12 @@ define i64 @same_exit_block_pre_inc_use1_reverse() {
 ; VF4IC4-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[TMP0]], i32 -12
 ; VF4IC4-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i8, ptr [[TMP18]], i32 -3
 ; VF4IC4-NEXT:    [[WIDE_LOAD13:%.*]] = load <4 x i8>, ptr [[TMP17]], align 1
-; VF4IC4-NEXT:    [[REVERSE14:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD13]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD1:%.*]] = load <4 x i8>, ptr [[TMP4]], align 1
-; VF4IC4-NEXT:    [[REVERSE2:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD1]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD3:%.*]] = load <4 x i8>, ptr [[TMP6]], align 1
-; VF4IC4-NEXT:    [[REVERSE4:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD3]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD5:%.*]] = load <4 x i8>, ptr [[TMP27]], align 1
+; VF4IC4-NEXT:    [[REVERSE14:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD13]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF4IC4-NEXT:    [[REVERSE2:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD1]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF4IC4-NEXT:    [[REVERSE4:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD3]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[REVERSE6:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD5]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[TMP25:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[OFFSET_IDX]]
 ; VF4IC4-NEXT:    [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[TMP25]], i32 0
@@ -680,12 +680,12 @@ define i64 @same_exit_block_pre_inc_use1_reverse() {
 ; VF4IC4-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[TMP25]], i32 -12
 ; VF4IC4-NEXT:    [[TMP42:%.*]] = getelementptr inbounds i8, ptr [[TMP16]], i32 -3
 ; VF4IC4-NEXT:    [[WIDE_LOAD14:%.*]] = load <4 x i8>, ptr [[TMP46]], align 1
-; VF4IC4-NEXT:    [[REVERSE15:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD14]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD9:%.*]] = load <4 x i8>, ptr [[TMP13]], align 1
-; VF4IC4-NEXT:    [[REVERSE10:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD9]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD11:%.*]] = load <4 x i8>, ptr [[TMP15]], align 1
-; VF4IC4-NEXT:    [[REVERSE12:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD11]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD15:%.*]] = load <4 x i8>, ptr [[TMP42]], align 1
+; VF4IC4-NEXT:    [[REVERSE15:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD14]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF4IC4-NEXT:    [[REVERSE10:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD9]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF4IC4-NEXT:    [[REVERSE12:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD11]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[REVERSE16:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD15]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[TMP21:%.*]] = icmp ne <4 x i8> [[REVERSE14]], [[REVERSE15]]
 ; VF4IC4-NEXT:    [[TMP19:%.*]] = icmp ne <4 x i8> [[REVERSE2]], [[REVERSE10]]
@@ -916,12 +916,12 @@ define i8 @same_exit_block_reverse_use_loaded_value() {
 ; VF4IC4-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i8, ptr [[TMP0]], i32 -12
 ; VF4IC4-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[TMP7]], i32 -3
 ; VF4IC4-NEXT:    [[WIDE_LOAD5:%.*]] = load <4 x i8>, ptr [[TMP8]], align 1
-; VF4IC4-NEXT:    [[REVERSE6:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD5]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD1:%.*]] = load <4 x i8>, ptr [[TMP4]], align 1
-; VF4IC4-NEXT:    [[REVERSE2:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD1]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD3:%.*]] = load <4 x i8>, ptr [[TMP6]], align 1
-; VF4IC4-NEXT:    [[REVERSE4:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD3]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x i8>, ptr [[TMP11]], align 1
+; VF4IC4-NEXT:    [[REVERSE6:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD5]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF4IC4-NEXT:    [[REVERSE2:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD1]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF4IC4-NEXT:    [[REVERSE4:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD3]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[REVERSE7:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD6]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[P2]], i64 [[OFFSET_IDX]]
 ; VF4IC4-NEXT:    [[TMP38:%.*]] = getelementptr inbounds i8, ptr [[TMP9]], i32 0
@@ -933,12 +933,12 @@ define i8 @same_exit_block_reverse_use_loaded_value() {
 ; VF4IC4-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[TMP9]], i32 -12
 ; VF4IC4-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[TMP16]], i32 -3
 ; VF4IC4-NEXT:    [[WIDE_LOAD13:%.*]] = load <4 x i8>, ptr [[TMP17]], align 1
-; VF4IC4-NEXT:    [[REVERSE14:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD13]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD9:%.*]] = load <4 x i8>, ptr [[TMP13]], align 1
-; VF4IC4-NEXT:    [[REVERSE10:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD9]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD11:%.*]] = load <4 x i8>, ptr [[TMP15]], align 1
-; VF4IC4-NEXT:    [[REVERSE12:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD11]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[WIDE_LOAD14:%.*]] = load <4 x i8>, ptr [[TMP18]], align 1
+; VF4IC4-NEXT:    [[REVERSE14:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD13]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF4IC4-NEXT:    [[REVERSE10:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD9]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; VF4IC4-NEXT:    [[REVERSE12:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD11]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[REVERSE15:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD14]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
 ; VF4IC4-NEXT:    [[TMP21:%.*]] = icmp ne <4 x i8> [[REVERSE6]], [[REVERSE14]]
 ; VF4IC4-NEXT:    [[TMP19:%.*]] = icmp ne <4 x i8> [[REVERSE2]], [[REVERSE10]]
diff --git a/llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge.ll b/llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge.ll
index 4a4bda254bf88..29224826c12de 100644
--- a/llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge.ll
+++ b/llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge.ll
@@ -1150,7 +1150,8 @@ define void @ptr_induction_remove_dead_recipe(ptr %start, ptr %end) {
 ; CHECK-NEXT:     CLONE ir<%ptr.iv.next> = getelementptr inbounds vp<[[PTR_IV]]>, ir<-1>
 ; CHECK-NEXT:     vp<[[VEC_PTR:%.+]]> = vector-end-pointer inbounds ir<%ptr.iv.next>, vp<[[VF]]>
 ; CHECK-NEXT:     WIDEN ir<%l> = load vp<[[VEC_PTR]]>
-; CHECK-NEXT:     WIDEN ir<%c.1> = icmp ne ir<%l>, ir<0>
+; CHECK-NEXT:     EMIT vp<%9> = reverse ir<%l>
+; CHECK-NEXT:     WIDEN ir<%c.1> = icmp ne vp<%9>, ir<0>
 ; CHECK-NEXT:   Successor(s): pred.store
 ; CHECK-EMPTY:
 ; CHECK-NEXT:   <xVFxUF> pred.store: {

>From 4b0798f691814ea3d7225c034e6ee71bf6167cdd Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Wed, 30 Jul 2025 01:22:31 -0700
Subject: [PATCH 02/10] Remove cost model for vp_reverse

---
 llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index 7bd32ffe18b64..c707fb110b10c 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -1523,12 +1523,6 @@ RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
                           cast<VectorType>(ICA.getArgTypes()[0]), {}, CostKind,
                           0, cast<VectorType>(ICA.getReturnType()));
   }
-  case Intrinsic::experimental_vp_reverse: {
-    return getShuffleCost(TTI::SK_Reverse,
-                          cast<VectorType>(ICA.getReturnType()),
-                          cast<VectorType>(ICA.getArgTypes()[0]), {}, CostKind,
-                          0, cast<VectorType>(ICA.getReturnType()));
-  }
   case Intrinsic::fptoui_sat:
   case Intrinsic::fptosi_sat: {
     InstructionCost Cost = 0;

>From 66c70811be282f2222fc1f78efe6dec840d96c94 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Wed, 30 Jul 2025 02:56:00 -0700
Subject: [PATCH 03/10] adjust recipe in tryToWidenMemory

---
 .../Transforms/Vectorize/LoopVectorize.cpp    | 38 ++++++++++++-------
 .../Transforms/Vectorize/VPRecipeBuilder.h    |  5 +--
 .../Transforms/Vectorize/VPlanTransforms.cpp  | 31 ---------------
 .../Transforms/Vectorize/VPlanTransforms.h    | 14 -------
 4 files changed, 27 insertions(+), 61 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 8bc4a12e0299d..cf1f65f81a85c 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7608,9 +7608,9 @@ void EpilogueVectorizerEpilogueLoop::printDebugTracesAtEnd() {
   });
 }
 
-VPWidenMemoryRecipe *
-VPRecipeBuilder::tryToWidenMemory(Instruction *I, ArrayRef<VPValue *> Operands,
-                                  VFRange &Range) {
+VPRecipeBase *VPRecipeBuilder::tryToWidenMemory(Instruction *I,
+                                                ArrayRef<VPValue *> Operands,
+                                                VFRange &Range) {
   assert((isa<LoadInst>(I) || isa<StoreInst>(I)) &&
          "Must be called with either a load or store");
 
@@ -7667,14 +7667,30 @@ VPRecipeBuilder::tryToWidenMemory(Instruction *I, ArrayRef<VPValue *> Operands,
     Builder.insert(VectorPtr);
     Ptr = VectorPtr;
   }
-  if (LoadInst *Load = dyn_cast<LoadInst>(I))
-    return new VPWidenLoadRecipe(*Load, Ptr, Mask, Consecutive, Reverse,
-                                 VPIRMetadata(*Load, LVer), I->getDebugLoc());
 
-  StoreInst *Store = cast<StoreInst>(I);
-  return new VPWidenStoreRecipe(*Store, Ptr, Operands[0], Mask, Consecutive,
+  if (auto *Load = dyn_cast<LoadInst>(I)) {
+    auto *LoadR =
+        new VPWidenLoadRecipe(*Load, Ptr, Mask, Consecutive, Reverse,
+                              VPIRMetadata(*Load, LVer), Load->getDebugLoc());
+    if (Reverse) {
+      Builder.insert(LoadR);
+      return new VPInstruction(VPInstruction::Reverse, {LoadR},
+                               LoadR->getDebugLoc());
+    }
+    return LoadR;
+  }
+
+  auto *Store = cast<StoreInst>(I);
+  VPValue *StoredVal = Operands[0];
+  if (Reverse) {
+    auto *ReverseR = new VPInstruction(VPInstruction::Reverse, {StoredVal},
+                                       Store->getDebugLoc());
+    Builder.insert(ReverseR);
+    StoredVal = ReverseR;
+  }
+  return new VPWidenStoreRecipe(*Store, Ptr, StoredVal, Mask, Consecutive,
                                 Reverse, VPIRMetadata(*Store, LVer),
-                                I->getDebugLoc());
+                                Store->getDebugLoc());
 }
 
 /// Creates a VPWidenIntOrFpInductionRecpipe for \p Phi. If needed, it will also
@@ -8747,10 +8763,6 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
   // bring the VPlan to its final state.
   // ---------------------------------------------------------------------------
 
-  // Adjust the result of reverse memory accesses.
-  VPlanTransforms::runPass(VPlanTransforms::adjustRecipesForReverseAccesses,
-                           *Plan);
-
   // Adjust the recipes for any inloop reductions.
   adjustRecipesForReductions(Plan, RecipeBuilder, Range.Start);
 
diff --git a/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h b/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h
index 41878e3c648e3..16eb8cffb47a0 100644
--- a/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h
+++ b/llvm/lib/Transforms/Vectorize/VPRecipeBuilder.h
@@ -96,9 +96,8 @@ class VPRecipeBuilder {
   /// Check if the load or store instruction \p I should widened for \p
   /// Range.Start and potentially masked. Such instructions are handled by a
   /// recipe that takes an additional VPInstruction for the mask.
-  VPWidenMemoryRecipe *tryToWidenMemory(Instruction *I,
-                                        ArrayRef<VPValue *> Operands,
-                                        VFRange &Range);
+  VPRecipeBase *tryToWidenMemory(Instruction *I, ArrayRef<VPValue *> Operands,
+                                 VFRange &Range);
 
   /// Check if an induction recipe should be constructed for \p Phi. If so build
   /// and return it. If not, return null.
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 8c8b2c54769e3..a1a871009754b 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -3820,34 +3820,3 @@ void VPlanTransforms::addBranchWeightToMiddleTerminator(
       MDB.createBranchWeights({1, VectorStep - 1}, /*IsExpected=*/false);
   MiddleTerm->addMetadata(LLVMContext::MD_prof, BranchWeights);
 }
-
-void VPlanTransforms::adjustRecipesForReverseAccesses(VPlan &Plan) {
-  if (Plan.hasScalarVFOnly())
-    return;
-
-  for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(
-           vp_depth_first_deep(Plan.getVectorLoopRegion()))) {
-    for (VPRecipeBase &R : *VPBB) {
-      auto *MemR = dyn_cast<VPWidenMemoryRecipe>(&R);
-      if (!MemR || !MemR->isReverse())
-        continue;
-
-      if (auto *L = dyn_cast<VPWidenLoadRecipe>(MemR)) {
-        auto *Reverse =
-            new VPInstruction(VPInstruction::Reverse, {L}, L->getDebugLoc());
-        Reverse->insertAfter(L);
-        L->replaceAllUsesWith(Reverse);
-        Reverse->setOperand(0, L);
-        continue;
-      }
-
-      if (auto *S = dyn_cast<VPWidenStoreRecipe>(MemR)) {
-        VPValue *StoredVal = S->getStoredValue();
-        auto *Reverse = new VPInstruction(VPInstruction::Reverse, {StoredVal},
-                                          S->getDebugLoc());
-        Reverse->insertBefore(S);
-        S->setOperand(1, Reverse);
-      }
-    }
-  }
-}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
index 173aa6066ab83..9cf62a35ae36b 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
@@ -347,20 +347,6 @@ struct VPlanTransforms {
   static void
   addBranchWeightToMiddleTerminator(VPlan &Plan, ElementCount VF,
                                     std::optional<unsigned> VScaleForTuning);
-
-  /// Add reverse recipes for reverse memory accesses.
-  /// For reverse loads, transform
-  ///   WIDEN ir<%L> = load vp<%addr>
-  /// into
-  ///   WIDEN ir<%L> = load vp<%addr>
-  ///   EMIT   vp<%RevL> = reverse ir<%L>
-  ///
-  /// For reverse stores, transform
-  ///   WIDEN store vp<%addr>, ir<%SVal>
-  /// into
-  ///   EMIT   vp<%RevS> = reverse ir<%SVal>
-  ///   WIDEN  store vp<%addr>, vp<%RevS>
-  static void adjustRecipesForReverseAccesses(VPlan &Plan);
 };
 
 } // namespace llvm

>From 72d3bc6f61d2534b4900561d1c5c595b98f68671 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Wed, 30 Jul 2025 03:05:33 -0700
Subject: [PATCH 04/10] remove braces

---
 llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 2a669bbd32ac7..9b310b9615858 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -943,9 +943,8 @@ Value *VPInstruction::generate(VPTransformState &State) {
   }
   case VPInstruction::ResumeForEpilogue:
     return State.get(getOperand(0), true);
-  case VPInstruction::Reverse: {
+  case VPInstruction::Reverse:
     return Builder.CreateVectorReverse(State.get(getOperand(0)), "reverse");
-  }
   default:
     llvm_unreachable("Unsupported opcode for instruction");
   }

>From d33737e25e83dd157cc3f88ffdb3338b6d6c82b1 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Thu, 31 Jul 2025 01:30:41 -0700
Subject: [PATCH 05/10] fix from comment

---
 llvm/lib/Transforms/Vectorize/LoopVectorize.cpp | 9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index cf1f65f81a85c..780b697058b79 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7682,12 +7682,9 @@ VPRecipeBase *VPRecipeBuilder::tryToWidenMemory(Instruction *I,
 
   auto *Store = cast<StoreInst>(I);
   VPValue *StoredVal = Operands[0];
-  if (Reverse) {
-    auto *ReverseR = new VPInstruction(VPInstruction::Reverse, {StoredVal},
-                                       Store->getDebugLoc());
-    Builder.insert(ReverseR);
-    StoredVal = ReverseR;
-  }
+  if (Reverse)
+    StoredVal = Builder.createNaryOp(VPInstruction::Reverse, StoredVal,
+                                     Store->getDebugLoc());
   return new VPWidenStoreRecipe(*Store, Ptr, StoredVal, Mask, Consecutive,
                                 Reverse, VPIRMetadata(*Store, LVer),
                                 Store->getDebugLoc());

>From a9d4ab04bc55cf1e64944fb5b18592c72ad097d9 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Thu, 31 Jul 2025 02:13:14 -0700
Subject: [PATCH 06/10] postpone the removing dead reverse operations

---
 llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index a1a871009754b..c5392a7119be0 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -2229,6 +2229,8 @@ static VPRecipeBase *optimizeMaskToEVL(VPValue *HeaderMask,
 
 static void convertToEVLReverse(VPlan &Plan, VPTypeAnalysis &TypeInfo,
                                 VPValue &AllOneMask, VPValue &EVL) {
+  SmallVector<VPRecipeBase *> ToRemove;
+
   for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(
            vp_depth_first_shallow(Plan.getVectorLoopRegion()->getEntry()))) {
     for (VPRecipeBase &R : make_early_inc_range(reverse(*VPBB))) {
@@ -2243,9 +2245,12 @@ static void convertToEVLReverse(VPlan &Plan, VPTypeAnalysis &TypeInfo,
           TypeInfo.inferScalarType(VPI), VPI->getDebugLoc());
       NewReverse->insertBefore(VPI);
       VPI->replaceAllUsesWith(NewReverse);
-      VPI->eraseFromParent();
+      ToRemove.push_back(VPI);
     }
   }
+
+  for (VPRecipeBase *R : ToRemove)
+    R->eraseFromParent();
 }
 
 /// Replace recipes with their EVL variants.

>From 61f053d3bd4631b31c8402247a012e2443696855 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Thu, 14 Aug 2025 00:28:16 -0700
Subject: [PATCH 07/10] fix TTI::CastContextHint

---
 .../lib/Transforms/Vectorize/VPlanRecipes.cpp | 28 +++++++++++++++----
 1 file changed, 22 insertions(+), 6 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 9b310b9615858..df19d07e86b0c 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -2238,21 +2238,37 @@ InstructionCost VPWidenCastRecipe::computeCost(ElementCount VF,
     return TTI::CastContextHint::Normal;
   };
 
+  using namespace llvm::VPlanPatternMatch;
   VPValue *Operand = getOperand(0);
   TTI::CastContextHint CCH = TTI::CastContextHint::None;
   // For Trunc/FPTrunc, get the context from the only user.
-  if ((Opcode == Instruction::Trunc || Opcode == Instruction::FPTrunc) &&
-      !hasMoreThanOneUniqueUser() && getNumUsers() > 0) {
-    if (auto *StoreRecipe = dyn_cast<VPRecipeBase>(*user_begin()))
-      CCH = ComputeCCH(StoreRecipe);
+  if (Opcode == Instruction::Trunc || Opcode == Instruction::FPTrunc) {
+    static auto GetOnlyUser = [](const VPSingleDefRecipe *R) -> VPRecipeBase * {
+      if (R->hasMoreThanOneUniqueUser() || R->getNumUsers() == 0)
+        return nullptr;
+      return dyn_cast<VPRecipeBase>(*R->user_begin());
+    };
+
+    if (VPRecipeBase *Recipe = GetOnlyUser(this)) {
+      if (match(Recipe, m_VPInstruction<VPInstruction::Reverse>(m_VPValue())))
+        Recipe = GetOnlyUser(cast<VPInstruction>(Recipe));
+      if (Recipe)
+        CCH = ComputeCCH(Recipe);
+    }
   }
   // For Z/Sext, get the context from the operand.
   else if (Opcode == Instruction::ZExt || Opcode == Instruction::SExt ||
            Opcode == Instruction::FPExt) {
     if (Operand->isLiveIn())
       CCH = TTI::CastContextHint::Normal;
-    else if (Operand->getDefiningRecipe())
-      CCH = ComputeCCH(Operand->getDefiningRecipe());
+    else if (auto *Recipe = Operand->getDefiningRecipe()) {
+      VPValue *ReverseOp;
+      if (match(Recipe,
+                m_VPInstruction<VPInstruction::Reverse>(m_VPValue(ReverseOp))))
+        Recipe = ReverseOp->getDefiningRecipe();
+      if (Recipe)
+        CCH = ComputeCCH(Recipe);
+    }
   }
 
   auto *SrcTy =

>From 69053f500a6c4d8079768c427637911a7b0bef66 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Mon, 25 Aug 2025 20:10:54 -0700
Subject: [PATCH 08/10] Change planContainsAdditionalSimplifications for licm

---
 llvm/lib/Transforms/Vectorize/LoopVectorize.cpp | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 780b697058b79..744fb9c202ed3 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -6939,6 +6939,23 @@ static bool planContainsAdditionalSimplifications(VPlan &Plan,
             cast<VPRecipeWithIRFlags>(R).getPredicate() !=
                 cast<CmpInst>(UI)->getPredicate())
           return true;
+
+        if (auto *MemR = dyn_cast<VPWidenMemoryRecipe>(&R)) {
+          if (MemR->isReverse()) {
+            // If the stored value of a reverse store is invariant, LICM will
+            // hoist the reverse operation to the preheader. In this case, the
+            // result of the VPlan-based cost model will diverge from that of
+            // the legacy model.
+            if (auto *StoreR = dyn_cast<VPWidenStoreRecipe>(MemR))
+              if (StoreR->getStoredValue()->isDefinedOutsideLoopRegions())
+                return true;
+
+            if (auto *StoreR = dyn_cast<VPWidenStoreEVLRecipe>(MemR))
+              if (StoreR->getStoredValue()->isDefinedOutsideLoopRegions())
+                return true;
+          }
+        }
+
         SeenInstrs.insert(UI);
       }
     }

>From 715a55ff55c3625ed1b3dcad5d37a36f6b75138c Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Mon, 25 Aug 2025 20:39:19 -0700
Subject: [PATCH 09/10] Remove lammbda static

---
 llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index df19d07e86b0c..4534614671dc8 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -2243,7 +2243,7 @@ InstructionCost VPWidenCastRecipe::computeCost(ElementCount VF,
   TTI::CastContextHint CCH = TTI::CastContextHint::None;
   // For Trunc/FPTrunc, get the context from the only user.
   if (Opcode == Instruction::Trunc || Opcode == Instruction::FPTrunc) {
-    static auto GetOnlyUser = [](const VPSingleDefRecipe *R) -> VPRecipeBase * {
+    auto GetOnlyUser = [](const VPSingleDefRecipe *R) -> VPRecipeBase * {
       if (R->hasMoreThanOneUniqueUser() || R->getNumUsers() == 0)
         return nullptr;
       return dyn_cast<VPRecipeBase>(*R->user_begin());

>From 7c2493d3a9bb658995291dcbb8ab07a4b36a1da4 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 26 Aug 2025 03:24:49 -0700
Subject: [PATCH 10/10] Change the way to convert reverse operation When EVL
 lowering.

---
 .../Transforms/Vectorize/VPlanPatternMatch.h  | 13 +++-
 .../Transforms/Vectorize/VPlanTransforms.cpp  | 76 ++++++++++++-------
 2 files changed, 61 insertions(+), 28 deletions(-)

diff --git a/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h b/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
index b33359c9bb0d6..40230964ec838 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanPatternMatch.h
@@ -245,7 +245,8 @@ struct Recipe_match {
     if constexpr (std::is_same<RecipeTy, VPScalarIVStepsRecipe>::value ||
                   std::is_same<RecipeTy, VPCanonicalIVPHIRecipe>::value ||
                   std::is_same<RecipeTy, VPDerivedIVRecipe>::value ||
-                  std::is_same<RecipeTy, VPWidenGEPRecipe>::value)
+                  std::is_same<RecipeTy, VPWidenGEPRecipe>::value ||
+                  std::is_same<RecipeTy, VPVectorEndPointerRecipe>::value)
       return DefR;
     else
       return DefR && DefR->getOpcode() == Opcode;
@@ -550,6 +551,16 @@ m_LogicalOr(const Op0_t &Op0, const Op1_t &Op1) {
   return m_Select(Op0, m_True(), Op1);
 }
 
+template <typename Op0_t, typename Op1_t>
+using VPVectorEndPointer_match =
+    Recipe_match<std::tuple<Op0_t, Op1_t>, 0, false, VPVectorEndPointerRecipe>;
+
+template <typename Op0_t, typename Op1_t>
+inline VPVectorEndPointer_match<Op0_t, Op1_t>
+m_VectorEndPointer(const Op0_t &Op0, const Op1_t &Op1) {
+  return VPVectorEndPointer_match<Op0_t, Op1_t>({Op0, Op1});
+}
+
 template <typename Op0_t, typename Op1_t, typename Op2_t>
 using VPScalarIVSteps_match = Recipe_match<std::tuple<Op0_t, Op1_t, Op2_t>, 0,
                                            false, VPScalarIVStepsRecipe>;
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index c5392a7119be0..ea9146091ee52 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -2151,6 +2151,22 @@ void VPlanTransforms::addActiveLaneMask(
   HeaderMask->eraseFromParent();
 }
 
+/// If \p R is a VPInstruction::Reverse, return a VPWidenIntrinsicRecipe
+/// for the vp.reverse intrinsic using \p EVL. Returns nullptr otherwise.
+static VPWidenIntrinsicRecipe *
+getEVLReverse(VPRecipeBase &R, VPTypeAnalysis &TypeInfo, VPValue &EVL) {
+  VPValue *ReversedVal;
+  if (!match(&R,
+             m_VPInstruction<VPInstruction::Reverse>(m_VPValue(ReversedVal))))
+    return nullptr;
+
+  auto *Reverse = cast<VPInstruction>(&R);
+  VPlan *Plan = Reverse->getParent()->getPlan();
+  return new VPWidenIntrinsicRecipe(
+      Intrinsic::experimental_vp_reverse, {ReversedVal, Plan->getTrue(), &EVL},
+      TypeInfo.inferScalarType(Reverse), Reverse->getDebugLoc());
+}
+
 /// Try to optimize a \p CurRecipe masked by \p HeaderMask to a corresponding
 /// EVL-based recipe without the header mask. Returns nullptr if no EVL-based
 /// recipe could be created.
@@ -2227,32 +2243,6 @@ static VPRecipeBase *optimizeMaskToEVL(VPValue *HeaderMask,
       .Default([&](VPRecipeBase *R) { return nullptr; });
 }
 
-static void convertToEVLReverse(VPlan &Plan, VPTypeAnalysis &TypeInfo,
-                                VPValue &AllOneMask, VPValue &EVL) {
-  SmallVector<VPRecipeBase *> ToRemove;
-
-  for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(
-           vp_depth_first_shallow(Plan.getVectorLoopRegion()->getEntry()))) {
-    for (VPRecipeBase &R : make_early_inc_range(reverse(*VPBB))) {
-      auto *VPI = dyn_cast<VPInstruction>(&R);
-      if (!VPI || VPI->getOpcode() != VPInstruction::Reverse)
-        continue;
-
-      SmallVector<VPValue *> Ops(VPI->operands());
-      Ops.append({&AllOneMask, &EVL});
-      auto *NewReverse = new VPWidenIntrinsicRecipe(
-          Intrinsic::experimental_vp_reverse, Ops,
-          TypeInfo.inferScalarType(VPI), VPI->getDebugLoc());
-      NewReverse->insertBefore(VPI);
-      VPI->replaceAllUsesWith(NewReverse);
-      ToRemove.push_back(VPI);
-    }
-  }
-
-  for (VPRecipeBase *R : ToRemove)
-    R->eraseFromParent();
-}
-
 /// Replace recipes with their EVL variants.
 static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
   VPTypeAnalysis TypeInfo(Plan);
@@ -2365,8 +2355,40 @@ static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
       CurVPV->replaceAllUsesWith(EVLRecipe->getVPSingleValue());
     }
     ToErase.push_back(CurRecipe);
+
+    // Convert general reverse operations on loaded values and stored values
+    // into vp.reverse, when the VPVectorEndPointerRecipe adjusting the access
+    // address uses EVL instead of VF.
+    // TODO: Extend conversion along the def-use/use-def chain, as reverse
+    // operations may be eliminated or moved in the future.
+    if (auto *MemR = dyn_cast<VPWidenMemoryRecipe>(EVLRecipe);
+        MemR && match(MemR->getAddr(),
+                      m_VectorEndPointer(m_VPValue(), m_Specific(&EVL)))) {
+      assert(MemR->isReverse() &&
+             "Only reverse access uses VPVectorEndPointerRecipe as address");
+      VPRecipeBase *Candidate = nullptr;
+      if (auto *LoadR = dyn_cast<VPWidenLoadEVLRecipe>(MemR)) {
+        assert(LoadR->getNumUsers() == 1 &&
+               "Unexpected user number of reverse load");
+        Candidate = cast<VPRecipeBase>(*LoadR->user_begin());
+      } else if (auto *StoreR = dyn_cast<VPWidenStoreEVLRecipe>(MemR)) {
+        VPValue *StoredVal = StoreR->getStoredValue();
+        // Skip if the stored value is not defined in the loop region.
+        if (StoredVal->isDefinedOutsideLoopRegions())
+          continue;
+        Candidate = StoredVal->getDefiningRecipe();
+      }
+      assert(Candidate && "Must have one reverse operation for reverse access");
+
+      VPWidenIntrinsicRecipe *NewReverse =
+          getEVLReverse(*Candidate, TypeInfo, EVL);
+      assert(NewReverse &&
+             "Unable to get an EVL reverse when tail folding by EVL");
+      NewReverse->insertBefore(Candidate);
+      cast<VPInstruction>(Candidate)->replaceAllUsesWith(NewReverse);
+      ToErase.push_back(Candidate);
+    }
   }
-  convertToEVLReverse(Plan, TypeInfo, *AllOneMask, EVL);
   // Remove dead EVL mask.
   if (EVLMask->getNumUsers() == 0)
     ToErase.push_back(EVLMask->getDefiningRecipe());



More information about the llvm-commits mailing list